Levi, P.T., Chopra, S., Pang, J.C., Holmes, A., Gajwani, M., Sassenberg, T.A., DeYoung, C.G. & Fornito, A. (2023). The effect

Levi, P.T., Chopra, S., Pang, J.C., Holmes, A., Gajwani, M., Sassenberg, T.A., DeYoung, C.G. & Fornito, A. (2023). The effect
of using group-averaged or individualized brain parcellations when investigating connectome dysfunction in psychosis.
Netzwerkneurowissenschaften, Advance publication. https://doi.org/10.1162/netn_a_00329.

The effect of using group-averaged or individualized brain

parcellations when investigating connectome dysfunction in

psychosis

Short title: Individualized parcellation and dysconnectivity in psychosis

Priscila T. Levi1, Sidhant Chopra2, James C. Pang1, Alexander Holmes1, Mehul Gajwani1,

Tyler A. Sassenberg3, Colin G. DeYoung3, Alex Fornito1

1. Turner Institute for Brain and Mental Health, Monash University, Melbourne, Australia

2. Abteilung für Psychologie, Yale Universität, New Haven, USA

3. Abteilung für Psychologie, University of Minnesota, Minnesota, USA

Abstrakt

Functional magnetic resonance imaging (fMRT) is widely used to investigate functional

coupling (FC) disturbances in a range of clinical disorders. Most analyses performed to date

have used group-based parcellations for defining regions of interest (ROIs), in which a single

parcellation is applied to each brain. This approach neglects individual differences in brain

functional organization and may inaccurately delineate the true borders of functional regions.

These inaccuracies could inflate or under-estimate group differences in case-control analyses.

Wir

investigated how

individual differences

in brain organization

influence group

comparisons of FC using psychosis as a case-study, drawing on fMRI data in 121 early

psychosis patients and 57 Kontrollen. We defined FC networks using either a group-based

parcellation or an individually-tailored variant of the same parcellation. Individualized

parcellations yielded more functionally homogeneous ROIs than group-based parcellations.

At individual connections level, case-control FC differences were widespread, but the group-

D
Ö
w
N
Ö
A
D
e
D

F
R
Ö
M
H

T
T

:
/
/

D
ich
R
e
C
T
.

ich
T
.

e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D

F
/

D
Ö

ich
/

1
0
1
1
6
2
N
e
N
_
A
_
0
0
3
2
9
2
1
4
2
0
1
4
N
e
N
_
A
_
0
0
3
2
9
P
D

B
j
G
u
e
S
T

Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3

based parcellation identified approximately 9% connections as dysfunctional than the

individualized parcellation. When considering differences at the level of functional networks,

the results from both parcellations converged. Our results suggest that a substantial fraction

of dysconnectivity previously observed in psychosis may be driven by the parcellation

method, rather than a pathophysiological process related to psychosis.

Author summary

Functional magnetic resonance imaging is widely used to map how brain network

dysfunction is affected by diverse diseases. A fundamental step in this work involved

defining specific brain regions, which act as network nodes in the analysis. Most research to

date has used a one-size-fits all approach, defining such regions on a template brain that is

then applied to individual people, which neglects the potential for variability in regional

borders and brain organization. Hier, we show that using an individualized approach to

region definition results in more valid area definitions and more conservative estimates of

brain network dysfunction in people with psychosis, indicating that at least some of the group

differences reported in the extant literature may be due to differences in regional definitions

rather than a consequence of the illness itself.

D
Ö
w
N
Ö
A
D
e
D

F
R
Ö
M
H

T
T

:
/
/

D
ich
R
e
C
T
.

ich
T
.

e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D

F
/

D
Ö

ich
/

1
0
1
1
6
2
N
e
N
_
A
_
0
0
3
2
9
2
1
4
2
0
1
4
N
e
N
_
A
_
0
0
3
2
9
P
D

B
j
G
u
e
S
T

Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3

Einführung

Psychosis is a neuropsychiatric condition that has long been thought to arise from

aberrant neural connectivity, or dysconnectivity, between neuronal populations (Andreasen et

al., 1998; Baker et al., 2019; Fornito et al., 2012; Nogovitsyn et al., 2022). Solch

dysconnectivity is often studied using a network-based approach (Fornito et al., 2016), mit

the brains of individuals being modelled as a collection of nodes, representing discrete brain

Regionen, connected by edges, representing inter-regional structural connectivity or functional

coupling (FC). This approach has revealed extensive FC disruptions in psychosis patients,

which are often characterized by a global decrease in FC upon which is superimposed more

network-specific increases and decreases (Argyelan et al., 2014; Baker et al., 2019; Chopra et

al., 2021; Fornito et al., 2012; Hummer et al., 2020; T. Li et al., 2017; Narr & Leaver, 2015;

Nogovitsyn et al., 2022; Tu et al., 2013). Jedoch, the reported findings have been

inconsistent, with reports of increased and decreased FC sometimes found within the same

network in different samples (Lynall et al., 2010; Moran et al., 2013; Whitfield-Gabrieli et

al., 2009; Woodward et al., 2011).

Some of these inconsistencies may be explained by methodological differences in

defining the nodes (brain regions of interest – ROIs) of the constructed brain networks, welche

is a fundamental step in network analysis that could affect the validity and interpretation of

subsequent results (Fornito et al., 2010, 2016; Zalesky, Fornito, Harding, et al., 2010). Jede

node should ideally represent a functionally specialized area with homogenous activity

(Eickhoff, Polizist, et al., 2018; Eickhoff, Yeo, et al., 2018), but there is no consensus on

the optimal way of parcellating the brain, meaning that investigators must rely on various

heuristic methods (Eickhoff, Polizist, et al., 2018; Eickhoff, Yeo, et al., 2018).

D
Ö
w
N
Ö
A
D
e
D

F
R
Ö
M
H

T
T

:
/
/

D
ich
R
e
C
T
.

ich
T
.

e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D

F
/

D
Ö

ich
/

1
0
1
1
6
2
N
e
N
_
A
_
0
0
3
2
9
2
1
4
2
0
1
4
N
e
N
_
A
_
0
0
3
2
9
P
D

B
j
G
u
e
S
T

Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3

The vast majority of studies in patients with psychosis have used a one-size-fits-all,

group-based approach in defining distinct ROIs. A parcellation using this approach is often

defined in a standardized coordinate space based on a sample average and then mapped to

individual participants via a spatial normalization procedure (Eickhoff, Yeo, et al., 2018).

This approach fails to consider interindividual variability in functional and anatomical brain

organization (Amunts et al., 2005; Mueller et al., 2013). Investigation of such variability with

resting-state fMRI (rsfMRI) has shown that, although most cortical areas can indeed be

robustly identified in every individual, their sizes and shapes vary across the population,

especially when using more fine-grained parcellation methods (Gordon et al., 2017).

Außerdem, the topographical locations of specific areas tend to shift between individuals,

sometimes across anatomical landmarks such as sulci and gyri (Gordon et al., 2017), welche

are often used as reference points in many standard parcellations (Fornito et al., 2016).

To better accommodate this individual variability, approaches have been developed to

derive individualized parcellations at either the level of canonical functional networks (S. Li

et al., 2016; Yeo et al., 2011) or cortical regions (Gordon et al., 2017; Kong et al., 2021).

These approaches have revealed that individual variability can considerably impact network

D
Ö
w
N
Ö
A
D
e
D

F
R
Ö
M
H

T
T

:
/
/

D
ich
R
e
C
T
.

ich
T
.

e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D

F
/

D
Ö

ich
/

1
0
1
1
6
2
N
e
N
_
A
_
0
0
3
2
9
2
1
4
2
0
1
4
N
e
N
_
A
_
0
0
3
2
9
P
D

Analysen. Zum Beispiel, regions assigned to one network in individual parcellations are often

assigned to a different network in the group average (Bijsterbosch et al., 2018), which could

impact FC analysis. The use of individually-tailored parcellations yields more functionally

homogeneous regions (Chong et al., 2017; Kong et al., 2021), and can improve predictions of

behaviour from FC (Kong et al., 2019). In der Tat, in healthy samples, individual differences in

B
j
G
u
e
S
T

Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3

the locations of functional regions, as represented by individualized parcellation, affect

predictions of fluid intelligence (Kong et al., 2019), life satisfaction (Bijsterbosch et al.,

2018), participant sex (Salehi et al., 2018), and performance in reading and working memory

tasks (Kong et al., 2021). Darüber hinaus, some estimates indicate that up to 62% of variance in

network edge strength (d.h., FC values) can be explained by the spatial variability of defined

Regionen (Bijsterbosch et al., 2018). These findings suggest that clinically important

relationships may be masked when using a group-based parcellation. Andererseits, diese

approaches present several challenges, such as establishing a correspondence between similar

regions in different people and accounting for differences in region size.

A particularly salient point in clinical studies, such as those of schizophrenia, is that

100

standard brain atlases have been derived from healthy participants, which may not adequately

101

capture the characteristic properties in the brain organization of patients (Glasser et al., 2016;

102

Schaefer et al., 2018). Patient-specific individual variability in functional organization can

103

influence the results of brain network analyses. In der Tat, one study has found that slight

104

displacements of a seed region in the thalamus can lead to significant differences in disorder-

105

related dysconnectivity (Welsh et al., 2010), emphasizing the importance of a valid and

106

consistent node definition.

107

One strategy to develop individualized parcellations is to adjust the borders of a

108

group-based template for each individual participant according to pre-defined functional

109

Kriterien. For instance, Chong et al. (Chong et al., 2017) developed a Bayesian algorithm

110

(called Group Prior Individualized Parcellation – GPIP) that uses a group-based template as a

111

prior to find an optimal corresponding parcellation on individual brains using individual FC

112

Daten. The group-based prior ensures that the same regions are mapped in each individual,

113

while updates to the individualized prior account for variability in the shape and size of each

114

parcellated region. Chong et. al. (Chong et al., 2017) have shown that this method yields

115

parcellated regions with increased intra-regional functional homogeneity and reduced

116

variance in connectivity strength between individuals (Chong et al., 2017). Hier, we used this

117

approach to compare FC disruptions observed in people with early psychosis using analyses

118

that rely on either a group-based or individualized parcellation. The parcellation algorithm

D
Ö
w
N
Ö
A
D
e
D

F
R
Ö
M
H

T
T

:
/
/

D
ich
R
e
C
T
.

ich
T
.

e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D

F
/

D
Ö

ich
/

1
0
1
1
6
2
N
e
N
_
A
_
0
0
3
2
9
2
1
4
2
0
1
4
N
e
N
_
A
_
0
0
3
2
9
P
D

B
j
G
u
e
S
T

Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3

119

(Chong et al., 2017) allowed us to match all brain regions across participants while

120

accounting for individual variability. Our analyses were conducted using the high-quality,

121

open-access data provided by the Human Connectome Project – Early Psychosis (Glasser et

122

al., 2013; HCP Early Psychosis 1.1 Data Release: Reference Manual HUMAN Connectome

123

PROJECT for Early Psychosis, 2021) (HCP-EP) resource. We tested two competing

124

hypotheses of how individual variability contributes to apparent FC disruptions in psychosis.

125

Under one hypothesis, a failure to consider individual variability may lead to erroneous

126

regional parcellations, adding noise to the analyses and reducing statistical power for

127

detecting valid group differences. In this case, we expect to see fewer differences between

128

patients and controls when using the group-based parcellation compared to individualized

129

parcellation. Alternativ, FC differences between groups may be largely driven by

130

variations in the underlying organization of each individual’s brain, rather than reflecting

131

specific differences in FC. In this case, we expect to see more differences using the group-

132

based parcellation.

133

134

135

136

137

138

139

140

D
Ö
w
N
Ö
A
D
e
D

F
R
Ö
M
H

T
T

:
/
/

D
ich
R
e
C
T
.

ich
T
.

e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D

F
/

D
Ö

ich
/

1
0
1
1
6
2
N
e
N
_
A
_
0
0
3
2
9
2
1
4
2
0
1
4
N
e
N
_
A
_
0
0
3
2
9
P
D

B
j
G
u
e
S
T

Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3

141

142

Ergebnisse

Hier, we present results obtained using group-level cortical parcellations provided by

143

Schaefer et al. (Schaefer et al., 2018) as the basis for our analysis, focusing on the 100-region

144

parcellation (s100). To ensure that our results are robust to the number of regions, Wir

145

repeated our analysis using the 200-region variant (s200) and after applying Global Signal

146

Regression (GSR). Results obtained using the s200 atlas, and results for both atlases after

147

GSR, can be found in the Supplementary Materials and are largely consistent with the

148

primary results reported in the following sections.

149

Spatial and functional properties of group-based vs individualized parcellation

150

Figur 1 shows examples of individualized parcellations generated for three

151

individuals compared with the original group-based s100 atlas. The individualized

152

parcellation algorithm preserved the same regions for every individual but shifted their

153

borders and changed their shapes and sizes to accommodate for individualized variations in

154

brain organization. In der Tat, on average, 42.56% (𝑆𝐷 = 2.37) of vertices were reallocated to a

155

different region as a result of the individualized parcellation algorithm, highlighting the

156

considerable variability of cortical functional organization between individuals. Figure 2a

157

shows the proportion of vertices that were relabelled in controls 𝑀(𝑆𝐷) = 43.28% (2.34)

158

and in patients 𝑀(𝑆𝐷) = 42.20% (2.31). The difference between the two groups was small

159

but statistically significant, following permutation testing ( 𝑝 = 0.004, 𝐻𝑒𝑑𝑔𝑒𝑠′𝑠 𝑔 =

160

0.465). Jedoch, at a regional level (figure 2b), no parcel showed significant differences in

161

the number of vertices relabelled between patients and controls (d.h., all 𝑝𝐹𝐷𝑅 >

162

0.05, corrected with the Benjamini and Hochberg method).

163

D
Ö
w
N
Ö
A
D
e
D

F
R
Ö
M
H

T
T

:
/
/

D
ich
R
e
C
T
.

ich
T
.

e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D

F
/

D
Ö

ich
/

1
0
1
1
6
2
N
e
N
_
A
_
0
0
3
2
9
2
1
4
2
0
1
4
N
e
N
_
A
_
0
0
3
2
9
P
D

B
j
G
u
e
S
T

Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3

D
Ö
w
N
Ö
A
D
e
D

F
R
Ö
M
H

T
T

:
/
/

D
ich
R
e
C
T
.

ich
T
.

e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D

F
/

D
Ö

ich
/

1
0
1
1
6
2
N
e
N
_
A
_
0
0
3
2
9
2
1
4
2
0
1
4
N
e
N
_
A
_
0
0
3
2
9
P
D

B
j
G
u
e
S
T

Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3

164

165

Figur 1. Differences in parcel boundaries between group-based and individualized

166

parcellation. The images show different parcellations overlayed on the inflated fsaverage5

167

template surface of the left hemisphere, mit 20,484 Eckpunkte. The top image shows the

168

group-based parcellation, which was used as a starting point for the individualized

169

parcellation algorithm. Colors correspond to the seven canonical functional networks that are

170

used to group parcels in the atlas (Yeo et al., 2011). The bottom three images show

171

individualized parcellations for three different subjects after 20 iterations of the GPIP

172

Algorithmus. The region shaded in orange corresponds to region 1 in the lateral prefrontal

173

cortex of the control network for all parcellations. The region shaded in red corresponds to

174

region 1 in the parietal lobe of the default mode network. The same regions are present in all

175

individuals, but their locations, sizes and shapes show considerable variability. DorsAttn –

176

dorsal attention network; SomMot – somatomotor network; Cont – control network; Default

177

– default mode network; Limbic – limbic network; SalVentAttn – salience/ventral attention

178

Netzwerk; Vis – visual network.

179

We next compared the average functional homogeneity of the group-based and individualized

180

parcellations. Functional homogeneity was measured out of sample, on functional scans from

181

run 2 with parcellations generated for scans from run 1. In controls, the mean homogeneity

182

War 0.364 (𝑆𝐷 = 0.09), Und 0.372 (𝑆𝐷 = 0.08) for the group-based and individualized

183

parcellations, jeweils. In patients, the mean homogeneity was 0.297 (𝑆𝐷 = 0.06) Und

184

0.304 (𝑆𝐷 = 0.06) for the group-based and individualized parcellations, jeweils (figure

185

2C). A two-way mixed ANOVA revealed that mean homogeneity was higher for the

186

individualized parcellation (𝐹(149) = 54.81, 𝑝 < 0.0001) and higher in controls compared 187 to patients (𝐹(149) = 30.91, 𝑝 < 0.0001), with no interaction between parcellation type and 188 diagnostic group (𝐹(149) = 0, 𝑝 = 0.898). Post-hoc analysis showed that individualized 189 parcellation resulted in greater homogeneity scores in patients (𝑡(103) = 5.64, 𝑝 < 0.0001) 190 and controls (𝑡(46) = 2.90, 𝑝 = 0.006). When comparing homogeneity scores for individual 191 parcels (figure 2d, e), 55 out of 85 regions showed significant differences in homogeneity 192 between parcellation approaches (i.e., 𝑝𝐹𝐷𝑅 < 0.05, corrected with the Benjamini and 193 Hochberg method). Moreover, both methods showed high reliability of homogeneity 194 estimates, as measured with the intraclass correlation coefficient (McGraw & Wong, 1996) l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / . / t / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t / . 195 (𝑟𝑔𝑟𝑜𝑢𝑝−𝑏𝑎𝑠𝑒𝑑 = 0.842, 𝑝 < 0.0001 𝑎𝑛𝑑 𝑟𝑖𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙𝑖𝑧𝑒𝑑 = 0.862, 𝑝 < 0.0001). To quantify 196 functional distinctions between parcels, we computed the mean Pearson’s correlation of 197 activity between each pair of vertices that were not allocated to the same region. We found 198 that the individualized parcellation (𝑀𝑐𝑜𝑟𝑟(𝑆𝐷) = 0.100 (0.066)) delineates parcels that are 199 slightly more functionally distinct than those in the group-based parcellation (𝑀𝑐𝑜𝑟𝑟(𝑆𝐷) = f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 200 0.102 (0.066)). Although small, the difference was statically significant (𝑡(165) = 201 14.0, 𝑝 < 0.001). 202 Homogeneity scores results were similar for s200 atlas with and without GSR 203 (Supplementary Materials figures 2 and 3). For the s100 atlas with GSR, differences in 204 homogeneity between groups and parcellation approach were similar to the main results. 205 However, there was a significant interaction effect between parcellation type and diagnosis 206 (𝐹(148) = 4.68, 𝑝 = 0.032) (See Supplementary Materials figure 1), such that homogeneity 207 scores in patients were more impacted by individualized parcellation than in controls. This 208 result suggests that, at this particular resolution, parcellation type may differentially affect FC 209 estimates in patients and controls only following the application of GSR. The reasons for this 210 sensitivity to parcellation scale and GSR are unclear. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / . / / t 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t . / f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 211 212 Figure 2 – Spatial and functional properties of group-based vs individualized 213 parcellations. Panel a shows the proportion of vertices relabelled by the individualized 214 parcellations for controls (𝑀(𝑆𝐷) = 0.433(0.023)) and for patients (𝑀(𝑆𝐷) = 215 0.422(0.023)). Panel b shows the average number of vertices relabelled in every parcel for 216 patients and controls. Panel c shows the distribution of homogeneity scores per subject. 217 Controls produced more homogenous parcels in both individualized (𝑀(𝑆𝐷) = 218 0.372(0.08)) and group-based parcellations (𝑀(𝑆𝐷) = 0.364(0.09)) than patients 219 (𝑖𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙𝑖𝑧𝑒𝑑 𝑀(𝑆𝐷) = 0.304(0.06)), (𝑔𝑟𝑜𝑢𝑝 − 𝑏𝑎𝑠𝑒𝑑 𝑀(𝑆𝐷) = 0.297(0.06)). Panel 220 d shows homogeneity scores for every parcel for group-based and individualized parcellation. 221 Light colored parcels in e represent parcels showing significant difference in homogeneity 222 scores, between parcellation approaches, for 𝑝𝐹𝐷𝑅 < 0.05. Homogeneity is displayed in 223 inflated surfaces with the group-based parcellation. 224 225 Unthresholded edge-level group differences in FC 226 Following exclusion of regions with poor signal (see Methods) the final networks 227 examined comprised 85 regions. The FC matrices resulting from both parcellation methods 228 were positively correlated, with correlations ranging between 0.679 and 0.898 (median = 229 0.794) across participants (Supplementary Materials figure 4a), indicating that the results 230 obtained with individualized and group parcellations are generally similar, although far from 231 identical. 232 Figure 3a shows the distribution of 𝑡-statistics across edges, comparing FC between 233 patients and controls estimated using either the group-based or individualized parcellation. 234 Both distributions have predominantly positive values, consistent with evidence of 235 widespread hypoconnectivity in patients compared to controls. The distribution for the group- 236 based approach is shifted further to the right, indicating that larger group differences are 237 detected with this method, on average. The difference in the means of the distributions was 238 statistically significant, as calculated with a Wilcoxon signed-rank test (𝑍 = 24.053 𝑝 < 239 0.0001). Figure 4 of the Supplementary Materials shows that most FC edges were positively 240 valued; as such, the significant FC reductions observed in patients result from patients having 241 lower positive FC rather than patients having stronger negative FC. Given the higher l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / . / t / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t / . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 242 functional homogeneity of the individualized parcellation, this result suggests that the group- 243 based parcellation overstates FC differences between patients and controls. Similar results 244 were obtained when looking at the effect size of the differences in edge strength between 245 patients and controls (Supplementary Materials figure 4), with the group-based parcellation 246 yielding higher effect size estimates than individualized parcellation, on average (𝑝 < 247 0.0001). 248 The 𝑡-matrices obtained using the group-based and individualized parcellations were 249 positively correlated (𝑟 = 0.76, 𝑝 < 0.0001), suggesting that the two approaches show 250 largely similar between-group FC differences. The effects of parcellation type were 251 consistent across the full extent of the 𝑡-distributions, as indicated by the shift function, 252 which compares differences between distributions at each decile. This analysis showed a 253 significantly higher value in every decile of the group-based parcellation, when compared to 254 the individualized parcellation, with the 95% CI never crossing zero (figure 3b). There was, 255 however, a more pronounced effect of parcellation type on edges associated with larger case- 256 control differences in FC relative to those with smaller case-control differences, as can be 257 seen by the greater shift observed in the right tail of the distribution relative to the left (figure 258 3b). This result implies that variations in parcellation type are more likely to influence the 259 edges that are significantly different between patients and controls. Furthermore, results 260 obtained using the s200 parcellations are in agreement with results obtained from the s100 261 parcellation (see Supplementary Materials figure 2). Following GSR, at both parcellation 262 scales, the mean t-values were similar, but the t-distribution for the individualized 263 parcellation was narrower than for the group-based parcellation. The shift function showed 264 that significant differences between parcellation approaches were mainly for edges with 265 positive t-values (see Supplementary Materials figures 1 and 3). l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / / / . t 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t / . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l 266 267 Figure 3 – Edge-specific case-control differences in FC depend on parcellation type. a 268 Distributions of 𝑡-values quantifying FC differences between patients and controls at each 269 edge and for each parcellation type. A positive t-value indicates a greater FC value in controls 270 than in patients. For reference, a p-value = 0.05 corresponds to a t-value = 1.65 uncorrected, 271 and t = 4.31 Bonferroni corrected. b Shift function (Rousselet et al., 2017) for the two t- 272 distributions. Each circle represents the difference between the borders of each decile of both 273 distributions as a function of the deciles in the group-based distribution. The bars represent 274 the 95% boot-strap confidence interval associated with the difference. 275 Thresholded edge-level group differences in FC 276 We used the Network Based Statistic (NBS) for inference on the edge-specific 𝑡-statistics 277 (Zalesky, Fornito, & Bullmore, 2010). The NBS identified a single connected component 278 with significant FC differences between patients and controls using both the group-based f / d o i / . / / t 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . / t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 279 (𝑝 < 0.0001) and individualized parcellations (𝑝 < 0.0001), for all primary test statistics 280 thresholds tested. Out of 3,570 possible connections, for a primary threshold equivalent to a 281 p-value = 0.05, the group-based and individualized parcellations resulted in components 282 comprising 2,877 edges and 2,672 edges respectively (figure 4a-b). Thus, the group-based 283 approach implicated approximately 7.7% more dysconnected edges. The binary edge 284 matrices defining these components were moderately correlated (𝑟𝑝ℎ𝑖 = 0.548, 𝑝 < 0.0001) 285 and both components had a total of 571 edges that differed from each other. There was also 286 some variation in the regional affiliation of the edges. For example, figure 4c-d show that the 287 insula has a high dysconnectivity degree in both group-based and individualized 288 parcellations, but that the former approach implicates more insula sub-regions. Furthermore, 289 the right medial prefrontal cortex shows a low degree in the individualized parcellation but 290 not in the group-based parcellation. The NBS was repeated with a primary test statistics 291 threshold equivalent to p-values = 0.01 and 0.001. For 𝑝 = 0.01, the component for 292 individualized parcellation comprised 1,786 edges and for group-based parcellation, 2,120. 293 For 𝑝 = 0.001, the component for individualized parcellation comprised 775 edges and for 294 group-based, 1,257 edges. Note that for all edges in these NBS networks, patients showed 295 reduced FC compared to controls. 296 297 Effects of variations in parcel size 298 A challenge of using individualized parcellations is that the ROIs can vary in size 299 across individuals, which may bias estimates of FC differences between groups. We therefore 300 examined changes in parcel size resulting from the individualization algorithm, as quantified 301 by the number of vertices in each parcel. On average, parcels changed by 50.7 (SD = 45.2) 302 vertices across patients and 52.0 (SD = 45.3) across controls, with no significant difference l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / / t / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / . t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 303 between the two groups, according to permutation testing (𝑝 = 0.104) (Supplementary 304 Materials figure 8a). There was also no significant difference in size difference between 305 patients and controls for any of the parcels, when corrected for multiple comparisons 306 following permutation statistics (i.e., all 𝑝𝐹𝐷𝑅 > 0.05). Four parcels had different sizes

307

between patients and controls, without correction for multiple comparisons (visual network

308

parcel 9 of the left hemisphere, 𝑝 = 0.023; somatomotor network parcel 1 of the left

309

hemisphere, 𝑝 = 0.026; limbic network parcel 1 in the orbital frontal cortex of the left

310

hemisphere, 𝑝 = 0.039, limbic network parcel 1 in the orbital frontal cortex of the right

311

hemisphere, 𝑝 = 0.048). We next correlated the differences in parcel size in individualized

312

parcellation between patients and controls with differences in node degree within the NBS

313

network and mean edge dysconnectivity, given by the mean 𝑡-value of edges attached to each

314

node for the case-control comparison (Supplementary Materials figure 8b-c). Neither

315

correlation was significant (𝑟 = 0.148, 𝑝𝑠𝑝𝑖𝑛 = 0.104 and 𝑟 = 0.133, 𝑝𝑠𝑝𝑖𝑛 = 0.127,

316

jeweils), suggesting that parcel size did not impact FC differences between patients and

317

controls in the individualized parcellation.

318

Network-level group differences in FC

319

Having demonstrated that the choice of a parcellation strategy can influence both

320

edge- and region-level inferences about FC disruptions in psychosis, we next examined

321

whether parcellation type affects the specific networks that are considered to be

322

dysfunctional. We therefore examined the proportion of edges within the NBS network that

323

fell within and between each of 7 canonical functional networks (Thomas Yeo et al., 2011).

324

Considering the raw number of affected edges across both parcellation approaches, Die

325

control network was the most impacted in patients with psychosis, with over 1,100

326

dysconnected edges, particularly those linking the control and somatomotor networks (figure

327

4e-f). By comparison, normalized counts, which is adjusted for the total number of possible

D
Ö
w
N
Ö
A
D
e
D

F
R
Ö
M
H

T
T

:
/
/

D
ich
R
e
C
T
.

ich
T
.

e
D
u
N
e
N
A
R
T
ich
C
e
–
P
D

F
/

D
Ö

ich
/

1
0
1
1
6
2
N
e
N
_
A
_
0
0
3
2
9
2
1
4
2
0
1
4
N
e
N
_
A
_
0
0
3
2
9
P
D

B
j
G
u
e
S
T

Ö
N
0
7
S
e
P
e
M
B
e
R
2
0
2
3

328

edges within or between pairs of networks, suggested a more equal and widespread

329

distribution of FC disruptions across networks. Both the raw count (𝑟 = 0.983, 𝑝 < 330 0.0001 ) and normalized matrices (𝑟 = 0.802, 𝑝 < 0.0001) were strongly correlated across 331 the two parcellation methods. These findings indicate that while parcellation method can 332 influence the specific edges that are identified as dysconnected, these edges generally fall 333 within or between the same canonical networks. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / . t / / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / . t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 334 335 Figure 4 – Edge-level regional and network-level case-control FC differences according 336 to parcellation type. Panels a and b show the specific edges comprising the NBS 337 components obtained with the group-based and individualized parcellations, respectively, 338 with nodes colored according to network affiliation and sized by degree. Edges are sized by 339 strength of dysconnectivity. Edges associated with a t-value < 3.5 are represented by grey 340 lines and those associated with a t-value ≥ 3.5 are represented in pink. The images were 341 created using the software BrainNet Viewer (Xia et al., 2013). Panels a, c, and e result from 342 the group-based parcellation. Panels c and d show the degree of each region in the NBS 343 component for the group and individualized parcellations, respectively. The left most triangle 344 of each matrix in panels e and f shows the total number of NBS component edges (raw 345 counts) falling within and between seven canonical networks. The right most triangles show 346 the same data normalized for network size, i.e. the total number of possible connection within 347 or between networks (normalized counts). DorsAttn – dorsal attention network; SomMot – 348 somatomotor network; Cont – control network; Default – default mode network; Limbic – 349 limbic network; SalVentAttn – salience/ventral attention network; Vis – visual network. 350 351 352 353 354 355 356 357 358 359 360 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / / t / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / t . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 361 362 Discussion Several studies have reported functional brain dysconnectivity in psychosis. A 363 fundamental step in such analyses involves defining a priori ROIs to serve as nodes in the 364 network analysis, which are typically derived from standard parcellation atlases generated 365 from a population or group average template. Here, we asked whether the failure of such an 366 approach to account for individual differences in brain functional organization can bias 367 estimates of case-control differences in FC. Standard methods could either result in an under- 368 estimation of the extent of network dysfunction (due to noisy FC estimation caused by 369 inaccurate ROI delineations) or an inflated estimate of the dysfunction (due to FC differences 370 being attributable to ROI misalignment), compared to when accounting for individual 371 differences in functional organization of the brain. Our findings indicate that group-based 372 parcellations might inflate estimates of FC differences in psychosis, especially at the edge- 373 level. Moreover, the use of individualized parcellations, while yielding a generally consistent 374 pattern of findings, leads to some different conclusions about the specific edges and regions 375 most affected by the disorder, although inferences at the network level were robust to 376 parcellation variations. Together, our findings suggest that the use of individualized 377 parcellations can impact findings of brain dysconnectivity in psychosis and, by extension, 378 other disorders. 379 Individualized parcellations yield more functionally homogeneous regions 380 The individualized parcellations resulted in nearly half (over 40%) of vertices being 381 assigned to regions that differed from the group-based atlas, as per prior work (Chong et al., 382 2017). This finding reiterates how group-based parcellations can result in a substantial 383 misspecification of regional borders in individuals and highlights the high degree of variance 384 present in the topographical organization of functional areas. Despite the high percentage of l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / / . t / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t / . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 385 vertices relabelled, FC matrices generated by both parcellations were highly correlated, 386 indicating the overall FC patterns seen with group-based parcellation are maintained with the 387 individualized parcellation. Note that with GPIP, correspondence between regions is 388 determined based on similarity in FC profiles rather than spatial location. As such, 389 corresponding regions can shift their spatial location from person to person (see Figure 1). 390 The higher functional homogeneity of the individualized parcellations supports its 391 improved validity, although the increment was small (2.4%), which is consistent with past 392 reports (Kong et al., 2021; Y. Li et al., 2022), increased homogeneity was seen in the 393 majority of parcels. Regional homogeneity was also marginally (2.3%) higher in controls 394 compared to patients. This differential improvement in homogeneity was expected, as the 395 starting point for the GPIP algorithm was the Schaefer atlas (Schaefer et al., 2018), which 396 was derived from a sample of people with no psychiatric disorders. Defining an initial group 397 atlas in patients would better account for differences in cortical functional organization 398 caused by psychosis. However, it would complicate comparisons between groups because of 399 the requirement to have consistently defined nodes in both patients and controls, which is one 400 of the challenges of using individualized parcellation. The interaction effect between 401 diagnosis and parcellation approach was not significant in most cases (apart from s100 with 402 GSR). This result indicates that individualized parcellations led to a similar improvement in 403 patients and controls. Since most case-control studies use data obtained in healthy individuals 404 to establish a normative benchmark for measures acquired in patients (Chopra et al., 2021; 405 Nabulsi et al., 2020; Nogovitsyn et al., 2022), we relied on the Schaefer parcellation in our 406 analysis. Future work could develop methods to better capture variations in functional 407 organization associated with psychosis. 408 Individualized parcellations lead to more conservative estimates of case-control FC 409 differences l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / t / . / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / t . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 410 Widespread decreases in FC in patients with psychosis were identified using both 411 parcellation approaches, highlighting that the dominant effect of both parcellations is 412 generally similar. However, the magnitude of the differences in FC was greater in the group- 413 based parcellation compared to individualized parcellation. Notably, the shift function 414 analysis indicated that differences between the two parcellation approaches were greater for 415 edges associated with large case-control differences. These edges are precisely the ones that 416 are most likely to be declared as statistically significant following the application of some 417 thresholding procedure. Accordingly, comparison of NBS results revealed a 7.7% reduction 418 in the size of the dysfunctional component identified using the group-based parcellation. 419 Given the higher functional homogeneity, and thus validity, of the individualized 420 parcellation, these results support the hypothesis that at least part of the group differences 421 identified in past studies in psychosis samples do not reflect actual differences in inter- 422 regional FC but instead result from inaccurate ROI boundaries caused by a failure to account 423 for individual differences in functional organization. These findings imply that individualized 424 parcellations can yield different estimates of FC differences in case-control studies, especially 425 when investigating FC changes at an edge-, or node-level. 426 Parcellation type affects FC differences in edges and regions, but not networks 427 While widespread decreases in FC were apparent in patients with psychosis using both 428 parcellation methods, the specific edges affected varied considerably. The NBS components 429 of both group-based and individualized parcellations showed differences in 571 edges (i.e., 430 19.8% of the total identified with the group-based parcellation). Examining the regions most 431 affected by quantifying the node degrees of the NBS components resulted in broadly similar 432 patterns, but there were some notable differences in location. For example, the right medial 433 frontal region accounts for 1.7% of dysconnectivity in the group-based and 2.3% in the 434 individualized parcellation. The left insula accounts for a slightly smaller percentage (6.5%) l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / . / t / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t / . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 435 of dysconnectivity in the group-based than in the individualized parcellation (6.7%). These 436 findings suggest that conclusions about the specific edges and regions affected by psychosis 437 can vary depending on the parcellation method used. In contrast, inferences at the network 438 level were largely consistent across the two parcellation approaches, indicating that coarse- 439 grained localizations of FC differences are robust to this methodological choice. This could 440 be attributed to network-level inference effectively reducing the dimensionality of the 441 analysis, minimizing the nuances of more fine-grained individual variations. Therefore, 442 studies looking at group differences in FC at a coarse, network level might not be impacted 443 by the use of individualized vs group-based parcellations. l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . 444 445 Limitations 446 To minimize the computational cost, we used fsaverage5, a surface mesh with a 447 relatively low number of vertices. Since GPIP parameters depend on the number of vertices 448 of the mesh, future work could investigate the impact of different surface mesh resolutions 449 and whether the differences observed here apply at different mesh resolution. 450 To facilitate comparison between subjects, the individualized parcellation algorithm 451 maintains the same number of regions for every subject and uses a parcellation derived in 452 healthy individuals as a starting point. This approach may mask differences in cortical 453 organization in patients, where regions may merge or split, resulting in a different number of 454 ROIs. However, generating separate parcellations in each group complicates comparisons 455 between groups. Resolving this challenge remains an open problem for the field. Moreover, 456 we only looked at cortical regions, due to the lack of methods available for individual 457 parcellation of subcortical structures. / / t e d u n e n a r t i c e - p d l f / d o i / / t / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / t . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 458 A proportion of patients in our sample were medicated, and recent evidence has 459 shown that anti-psychotic medication can impact FC, even after only 3 months of use 460 (Chopra et al., 2021). However, given that most samples examined in past research are also 461 medicated, our sample is directly comparable to the broader literature. Similarly, the study 462 included more patients than controls and future work could benefit from a balanced sample 463 size. We also emphasize that this study is not focused on identifying the specific nature of FC 464 disturbances associated with psychosis but instead concentrates on how parcellation type 465 affects FC differences in the same patients. In this context, medication exposure was constant 466 across our main contrast of interest (parcellation type), meaning that it cannot explain the 467 differences that we focus on here. The same reasoning applies to the clinical heterogeneity of 468 the patient sample, which comprised people diagnosed with both affective and non-affective 469 psychoses. Future work could use individualized parcellations to delineate FC differences 470 more precisely between distinct patient subgroups. 471 We have focused here on how the use of individualized vs group-based parcellations 472 affects group differences in FC. A separate question concerns whether parcellation type also 473 affects correlations with behavioural or clinical variables. Several studies have shown that 474 individualized parcellations yield FC estimates that are marginally more correlated with 475 various forms of behaviour, including psychopathological ratings (Bijsterbosch et al., 2018; 476 Kong et al., 2019, 2021). A useful direction for future work could involve investigating 477 whether individualized parcellation improves prediction of clinically meaningful outcomes. 478 479 Conclusion 480 Our findings indicate that traditional reliance on group-based parcellations may inflate case- 481 control differences in FC at a fine-grained level. The use of individualized parcellations can l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / / / t . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / t . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 482 yield a more conservative understanding of brain network disruptions in psychotic and 483 possibly other disorders. However, it does not greatly impact case-control differences in 484 network level analyses. 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / . t / / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . / t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 500 Methods 501 Study participants 502 All data for this study were collected as part of the Human Connectome Project – 503 Early Psychosis (HCP-EP) study, which is an open-access collection aiming to generate high- 504 quality imaging data in early psychosis patients and healthy controls (HCP Early Psychosis 505 1.1 Data Release: Reference Manual HUMAN Connectome PROJECT for Early Psychosis, 506 2021). This study includes high-resolution structural and functional Magnetic Resonance 507 Image (MRI) data from 121 patients with early psychosis (74 males) and 57 healthy 508 individuals (37 males). Demographic information is provided in Table 1. Data collection by 509 HCP-EP has been approved by the Partners Healthcare Human Research Committee/IRB, 510 and comply with the regulations set forth by the Declaration of Helsinki (Lewandowski et al., 511 2020). 512 The patient group was comprised of outpatients with psychosis, meeting criteria for 513 affective or non-affective psychosis, according to the DSM-5, who were within the first five 514 years of onset of symptoms. Patients were recruited by four clinical programs: Beth Israel 515 Deaconess Medical Center (BMH) – Massachusetts Mental Health Center (BIDMC-MMHC), 516 Prevention of and Recovery from Early Psychosis (PREP) Program; Indiana University 517 Psychotic Disorders Program, Prevention and Recovery for Early Psychosis (PARC); the 518 McLean Hospital, McLean On Track; and Massachusetts General Hospital, First Episode and 519 Early Psychosis Program (FEPP) (HCP Early Psychosis 1.1 Data Release: Reference Manual 520 HUMAN Connectome PROJECT for Early Psychosis, 2021). Imaging took place in three of 521 these sites. 522 The control group included volunteers that did not present with anxiety disorders 523 and/or psychotic disorders, had no first-degree relative with schizophrenia spectrum disorder, l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / / t / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / t . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 524 were not taking psychiatric medication at the time of the study, and had never been 525 hospitalized for psychiatric reasons. All participants were aged between 16 and 35 years old 526 (mean = 23, SD = ±3.9) at the time of the study (Table 1). A total of 11 subjects were 527 excluded due to poor data quality, as detailed below, leaving a final sample of 55 (36 male) 528 controls and 112 (67 male) patients. 529 Table 1. Demographic details Age Sex Control AP NAP 24.7 (4.1) 24.2 (4.3) 22.1 (3.3) 36M; 19F 7M; 19F 60M; 26F Antipsychotic -- 1.5 (0 – 54) 11.5 (0 – 56) exposure (months) NIH cognition 113.5(8.8) 108.9 (7.8) 98.2 (13.0) PANSS total score UI Scan site BMH McLean -- 23 26 6 40.7 (12.6) 48.8 (16.7) 7 9 10 48 30 8 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / / t / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t / . 530 AP – affective psychosis; NAP – non-affective psychosis; PANSS – Positive and Negative 531 Syndrome Scale; IU – Indiana University; BMH – Beth Israel Deaconess Medical 532 Center; Cont – healthy controls; F – females; M – males; age is given as mean (SD) in years 533 at the time of their first interview; antipsychotic exposure is given as median (range) in 534 months at the time of their first interview; PANSS total score is given as mean (SD); NIH 535 cognition is given as the mean (SD) of cognitive composite score, unadjusted for age, f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 536 assessed by the NIH Toolbox. 537 Data Acquisition 538 The participants recruited from four locations were scanned at three sites: BMH; 539 Indiana University; and McLean Hospital, using Siemens MAGNETOM Prisma 3T scanners. 540 The acquisition parameters between the three sites were harmonized and followed the widely 541 used HCP protocol (Demro et al., 2021; HCP Early Psychosis 1.1 Data Release: Reference 542 Manual HUMAN Connectome PROJECT for Early Psychosis, 2021). The project collected 543 whole brain T1-weighted MRI (T1w), T2-weighted MRI (T2w), diffusion MRI, spin echo 544 field maps with Anterior to Posterior (AP) and Posterior to Anterior (PA) phase encoding 545 (PE) directions - and four resting-state functional MRI (rsfMRI) sessions. The current study 546 uses the T1w and T2w images, the spin echo field maps, and the first two runs of the rsfMRI 547 scans. A 32-channel head coil was used at BMH and Indiana University. A 64-channel head 548 and neck coil, with neck channels turned off was used at McLean Hospital. Real-time image 549 reconstruction and processing was performed for quality control and scans with detectable 550 problems were repeated (HCP Early Psychosis 1.1 Data Release: Reference Manual 551 HUMAN Connectome PROJECT for Early Psychosis, 2021). 552 Structural MRI acquisition parameters 553 Acquisition parameters followed HCP standards. T1w images were obtained using a 554 magnetization-prepared rapid gradient-echo (MPRAGE), with 0.8 mm isotropic spatial 555 resolution echo time (TE) = 2.22 ms, repetition time (TR) = 2400 ms, and field of view (FoV) 556 = 256 mm. T2w images were acquired following a 3D-SPACE sequence, with 0.8 mm 557 isotropic spatial resolution, TE = 563 ms, TR = 33200 ms, and FoV = 256 mm (HCP Early 558 Psychosis 1.1 Data Release: Reference Manual HUMAN Connectome PROJECT for Early 559 Psychosis, 2021). 560 Functional MRI acquisition parameters l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / t / / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t . / f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 561 The present study mainly utilized the first rsfMRI run (with anterior to posterior phase 562 encoding). The second run (with posterior to anterior phase encoding) was used to validate 563 the parcellation with out-of-sample analysis of within-parcel homogeneity. Scans were 564 acquired for a length of 6.5 minutes, resulting in a total of 420 volumes; the first 10 volumes 565 were removed prior to the dataset release. Images have an isotropic spatial resolution of 2 566 mm, TE = 37 ms, TR = 800 ms, and FoV = 208 mm. A multi-band acceleration factor of 8 567 was used to improve spatial and temporal resolution (HCP Early Psychosis 1.1 Data Release: 568 Reference Manual HUMAN Connectome PROJECT for Early Psychosis, 2021). 569 Structural and Functional Image Analysis 570 Raw Image Quality Control 571 All analyses were done on the MASSIVE high-performance computing facility 572 (Goscinski et al., 2014). 573 Raw structural and functional images were first visually inspected for large artefacts 574 and distortions. Images were then put through an automated quality control pipeline 575 (MRIQC) (Esteban et al., 2017) which computes 15 image quality metrics for each scan with 576 the purposes of identifying outliers warranting closer inspection. At this stage, three subjects 577 were excluded for missing or unusable structural images. 578 Head motion is a major source of noise in fMRI signals. Its effects remain present 579 even after volume realignment and can introduce systematic bias in case-control studies when 580 not strictly controlled (Parkes et al., 2018; Power et al., 2012). Head motion during the fMRI 581 scan was estimated using frame-wise displacement (FD), which is a summary measure of the 582 movement of the head from one volume to the next (Parkes et al., 2018). For each scan, FD 583 was calculated according to the method described by Jenkinson et al. (Jenkinson et al., 2002) 584 and the resulting trace was band-pass filtered and down sampled to account for the high l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / / . t / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t . / f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 585 sampling rate of the multiband fMRI acquisition (Power et al., 2019). Subjects were excluded 586 if they met at least one of the following stringent exclusion criteria: scans had a mean filtered 587 FD greater than 0.25 mm; more than 20% of frames were displaced by more than 0.2 mm; or 588 any frame was displaced by more than 5 mm. These criteria have previously been shown to 589 effectively mitigate motion-related contamination in fMRI connectivity analyses (Parkes et 590 al., 2018). In total, 11 subjects (2 controls) were excluded for excessive head movement in 591 the scanner. 592 Image Preprocessing 593 T1w images were processed using FreeSurfer version 6.0.1 (Dale et al., 1999) to 594 generate cortical surface models for each participant. Surfaces were visually examined for 595 inaccuracies and distortions. The fMRI data were processed according to the Minimal 596 Preprocessing Pipeline for HCP data (Glasser et al., 2013). The pipeline adapts steps from 597 FMRIB Software Library (FSL) and FreeSurfer to account for greater spatial and temporal 598 resolution and HCP-like distortions resulting from acquisition choices such as multiband 599 acceleration (Glasser et al., 2013). Briefly, images were skull stripped by the brain extraction 600 tool (BET) (Smith, 2002) of FSL, which removes non-brain matter from the image. Skull 601 stripped T1w, T2w, and fMRI were aligned using FMRIB’s Linear Image Registration Tool 602 (FLIRT) (Jenkinson et al., 2002; Jenkinson & Smith, 2001). Spin Echo EPI field maps with 603 opposite phase encoding directions were used to estimate spatial distortion caused by 604 magnetic field inhomogeneities, with corrections applied using FSL’s “topup” (Andersson et 605 al., 2003) and FLIRT. This process was fine-tuned and optimized using FreeSurfer’s 606 BBRegister (Greve & Fischl, 2009). Furthermore, bias field correction was performed on 607 structural images to remove gradients of voxel intensity differences, following the HCP 608 pipeline (Glasser et al., 2013). The fMRI volumes were realigned to the first volume for each 609 participant using FLIRT. The fMRI data were then co-registered to their structural image, and l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / / / t . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . t / f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 610 the structural image was non-linearly normalized into standard Montreal Neurological 611 Institute (MNI) ICBM152 space (Grabner et al., 2006) using FLIRT and FMRIB’s nonlinear 612 image registration tool (FNIRT) (Andersson et al., 2010). The resulting transform was then 613 applied to the functional data. 614 fMRI Denoising 615 The functional data were denoised using Independent Component Analysis (ICA)- 616 based X-noiseifier (FIX), which decomposes the data into spatially independent components 617 and uses machine learning to label each resulting component as either signal or noise 618 (Griffanti et al., 2014; Salimi-Khorshidi et al., 2014). The preprocessed fMRI timeseries were 619 then regressed against the estimated noise component signals and the residuals were retained 620 for further analysis. Component decomposition was performed using Multivariate 621 Exploratory Linear Optimized Decomposition into Independent Components (MELODIC) 622 (Griffanti et al., 2014; Salimi-Khorshidi et al., 2014). HCP’s training set – HCP_hp2000, 623 which includes pre-trained weights to classify independent components, was used as the 624 training set for the algorithm. A temporal high-pass filter (2000s Full Width Half Maximum) 625 was applied to remove low-frequency signal drifts, as recommended by the HCP 626 preprocessing guideline (Glasser et al., 2013). Following HCP’s guidelines (Glasser et al., 627 2013), a lenient threshold component labelling in FIX was used (th=10), regressing out the 628 noise components while controlling for the signal components. The accuracy of the labels 629 was manually verified. The analyses were repeated after applying Global Signal Regression 630 (GSR), which removes widespread signal fluctuations associated with respiratory variations 631 (Aquino et al., 2020; Power et al., 2017) (see Supplemental Material). 632 Surface Registration l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / / t / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t / . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 633 The processed images in MNI volume space were resampled to each individual’s 634 cortical surface, as generated by FreeSurfer, and then registered to the fsaverage5 template 635 using a surface-based registration algorithm (Dale et al., 1999; Fischl, 2012). fsaverage5 is a 636 standard template generated by FreeSurfer, the resulting surface mesh comprises a total of 637 20,484 vertices. 638 Parcellations 639 We used group parcellations provided by Schaefer et al. (Schaefer et al., 2018) as the 640 basis for our analysis, as this parcellation is widely used and has shown superior functional 641 homogeneity compared to other leading approaches (Schaefer et al., 2018). Our study 642 focused on the 100-region parcellation, organized into 7 networks (s100) but we repeated the 643 analyses using the 200-region variant to check the robustness of the results (see 644 Supplementary Materials). Regions were screened for low BOLD signal intensity, with a 645 method adapted from Brown et. al. (Brown et al., 2019). Specifically, we found the elbow of 646 the BOLD signal distribution, given by the largest decrease in pair-wise differences of the 647 mean BOLD signal of each region. This was used as a cut-off for signal dropout and regions 648 with lower signal than the cut-off were considered to have signal dropout. Regions that were 649 found to have signal dropout in over 5% of subjects were excluded before analysis. For the 650 s100 atlas, 15 regions were excluded; for the s200 atlas, 16 regions were excluded from 651 further analysis. 652 To derive individually-tailored parcellations, we used the Group Prior Individualized 653 Parcellation (GPIP) model (Chong et al., 2017), which relies on a Bayesian formulation with 654 two priors: one based on group FC and one that drives individualized parcel boundaries. The 655 former uses a group sparsity constraint to represent FC between parcels, which allows the 656 model to maintain comparability between subjects. The latter uses a Markov Random Field in l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / . t / / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / . t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 657 the form of a Potts model to label the set of parcels and maximize the FC homogeneity within 658 each parcel based on individual data. This model allows for comparability between subjects, 659 as it maintains the same areas and labels for every individual while capturing the variability 660 in the shape and size of each parcel to best estimate each subject’s functional regions. 661 Individualized parcel borders were optimised across 20 iterations, starting with the group- 662 based Schaefer atlas and iteratively alternating between updating individual borders and the 663 group FC prior. Further details are provided in Chong et al. (Chong et al., 2017). The 664 algorithm was applied to patients and controls together. 665 For both group-based and individualized parcellations, mean timeseries were 666 extracted for each region in the s100 and s200 atlases using each individual’s spatially 667 normalized and denoised functional data. Product-moment correlations were then estimated 668 for every pair of regional time series to generate FC matrices. We only consider cortical areas 669 here as, to our knowledge, methods for developing individualized parcellations for 670 subcortical and cerebellar regions have not yet been developed. 671 Parcellation homogeneity and variability 672 To investigate the differences in parcels between the two parcellation approaches, we 673 computed how many vertices were reassigned to a different parcel after applying GPIP. We 674 then compared the number of vertices relabelled between patients and controls at a ROI and 675 whole-brain levels. All between-group statistical analyses were evaluated using permutation- 676 based inference, with 5000 permutations, unless otherwise indicated. Statistically significant 677 effects for ROI-level analysis were identified using an FDR-corrected (Benjamini & 678 Hochberg, 1995) threshold of 𝑝𝐹𝐷𝑅 < 0.05, two-tailed. 679 We compared the within-parcel functional homogeneity of the group-based and 680 individualized parcellations as per prior work (Chong et al., 2017; Schaefer et al., 2018). We l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / / t . / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . / t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 681 calculated the average FC between all pairs of vertices in a given parcel 𝑖, denoted 𝐹𝐶𝑖. Then, 682 parcellation homogeneity 𝐻 was normalised by parcel size as follows: 683 𝐻 = 𝑛 𝑖=1 ∑ 𝐹𝐶𝑖 × 𝑁𝑉𝑖 𝑛 ∑ 𝑁𝑉𝑖 𝑖=1 684 where 𝑛 is the total number of parcels in the parcellation and 𝑁𝑉 is the number of vertices in 685 the 𝑖𝑡ℎ parcel. This analysis was done out of sample, on functional scans from the second run 686 (PE=PA) with parcellations generated for scans from the first run (PE=AP). 687 To measure intra-subject reliability, we also computed homogeneity scores in the first 688 run and compared these results between parcellation approaches, using the intraclass 689 correlation coefficient. 690 Case-control differences in inter-regional functional coupling 691 We assessed how parcellation type influences FC differences between patients with 692 psychosis and healthy controls in three ways. First, we examined the distribution of 693 unthresholded t-statistics obtained at each edge using a general linear model to quantify mean 694 differences between patients and controls groups. This and all subsequent analyses are 695 controlled for age, sex, test site, and mean FD. The contrast was specified such that a larger t- 696 statistic indicated lower FC in patients, compared to controls. To compare the similarity of 697 the symmetric t-matrices, we vectorized their upper triangles and computed their Spearman 698 correlation. The effect of parcellation type was evaluated using a shift function test on these 699 distributions (Rousselet et al., 2017) to evaluate whether differences between parcellations 700 were restricted to specific quantiles of the 𝑡-statistic distributions (rather than just comparing 701 the means of these distributions). The shift function computes the difference in value of the 9 702 deciles of the distributions. For inference, it computes the 95% CI associated with each decile 703 difference, based on a bootstrap estimation of the standard error of each decile, controlling l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / . t / / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / . t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 704 for multiple comparisons, via the Hochberg’s method. This analysis thus allowed us to 705 determine whether parcellation type preferentially affected results for edges that showed 706 small, moderate, or large case-control differences. 707 Second, we compared thresholded results obtained with the Network Based Statistic 708 (NBS) (Zalesky, Fornito, & Bullmore, 2010). NBS is an adaptation of cluster-based statistics 709 for network data. A primary threshold of 𝑝 = 0.05, uncorrected, was applied to the matrix of 710 𝑡-statistics obtained using the general linear model described above. Results were repeated 711 with a threshold p = 0.01 and 0.001. The sizes of the connected components of the resulting 712 network (in terms of number of edges) were then estimated. In this context, the connected 713 components represent sets of nodes through which a path can be found via supra-threshold 714 edges. The group labels (patients and controls) were permuted 5000 times and the previous 715 steps were repeated. At each step, the size of the largest connected component was retained, 716 resulting in an empirical distribution of maximal component sizes under the null hypothesis. 717 The fraction of null values that exceeded the observed component sizes corresponds to a 718 family-wise corrected 𝑝-value for each component. The test was repeated with different 719 FWER corrected p-values = 0.05, 0.01, and 0.001, all resulting in the same connected 720 component. By performing inference at the level of connected components rather than 721 individual edges, the NBS results in greater statistical power than traditional mass univariate 722 thresholding methods (Zalesky, Fornito, & Bullmore, 2010). This analysis was repeated for 723 each parcellation type (i.e., group-based and individualized) and scale (i.e., s100 and s200). 724 Differences between significant component sizes observed using the two parcellation 725 methods were then estimated and evaluated with respect to the differences between null 726 component sizes estimated for the two approaches. 727 We calculated changes in parcel size between parcellation approaches for patients and 728 controls. We compared parcel size difference with a two-sample t-test between patients and l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / t / . / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . t / f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 729 controls. To understand how parcel size impacted FC measures, we calculated the Spearman 730 rho’s correlation between the t-values for parcel size and mean dysconnectivity per parcel 731 and degree of dysconnectivity. p-values were calculated with a spin permutation test, with 732 5000 permutations (Alexander-Bloch et al., 2018). 733 Finally, we examined how parcellation type affects case-control differences at the 734 level of 7 canonical networks. We considered the control network; the default mode network; 735 the dorsal attention network; the limbic network; the salience/ventral attention network; the 736 somatomotor network; and the visual network using the seven Yeo network assignments 737 associated with the s100 and s200 atlases (Yeo et al., 2011). Specifically, we quantified the 738 number of edges in the significant NBS component that fell within and between these seven 739 networks. We examined both raw edge counts and counts normalized for the size of each 740 network/network pair and quantified the correlation between the resulting network-level 741 matrices obtained for each parcellation type. 742 Code used for analysis and image generation can be found on-line at 743 https://github.com/NSBLab/individualised_parc_psychosis and code for individualized 744 parcellation can be acquired on-line at https://neuroimageusc.github.io/GPIP 745 746 747 748 749 750 751 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / t . / / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / . t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 752 Reference 753 Alexander-Bloch, A. F., Shou, H., Liu, S., Satterthwaite, T. D., Glahn, D. C., Shinohara, R. T., Vandekar, 754 755 756 S. N., & Raznahan, A. (2018). On testing for spatial correspondence between maps of human brain structure and function. NeuroImage, 178, 540–551. https://doi.org/10.1016/j.neuroimage.2018.05.070 757 Amunts, K., Kedo, O., Kindler, M., Pieperhoff, P., Mohlberg, H., Shah, N. J., Habel, U., Schneider, F., & 758 759 760 Zilles, K. (2005). Cytoarchitectonic mapping of the human amygdala, hippocampal region and entorhinal cortex: Intersubject variability and probability maps. Anatomy and Embryology, 210(5–6), 343–352. https://doi.org/10.1007/s00429-005-0025-5 761 Andersson, J. L. R., Jenkinson, M., & Smith, S. (2010). FMRIB Analysis Group Technical Reports. 762 https://www.fmrib.ox.ac.uk/datasets/techrep/ 763 Andersson, J. L. R., Skare, S., & Ashburner, J. (2003). How to correct susceptibility distortions in spin- 764 765 echo echo-planar images: Application to diffusion tensor imaging. NeuroImage, 20(2), 870– 888. https://doi.org/10.1016/S1053-8119(03)00336-7 766 Andreasen, N. C., Paradiso, S., & O’Leary, D. S. (1998). “Cognitive dysmetria” as an integrative theory 767 768 of schizophrenia: A dysfunction in cortical-subcortical-cerebellar circuitry? Schizophrenia Bulletin, 24(2), 203–218. https://doi.org/10.1093/oxfordjournals.schbul.a033321 769 Aquino, K. M., Fulcher, B. D., Parkes, L., Sabaroedin, K., & Fornito, A. (2020). Identifying and 770 771 772 removing widespread signal deflections from fMRI data: Rethinking the global signal regression problem. NeuroImage, 212, 116614. https://doi.org/10.1016/j.neuroimage.2020.116614 773 Argyelan, M., Ikuta, T., Derosse, P., Braga, R. J., Burdick, K. E., John, M., Kingsley, P. B., Malhotra, A. 774 775 776 K., & Szeszko, P. R. (2014). Resting-state fMRI connectivity impairment in schizophrenia and bipolar disorder. Schizophrenia Bulletin, 40(1), 100–110. https://doi.org/10.1093/schbul/sbt092 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / / / t . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / t . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 777 Baker, J. T., Dillon, D. G., Patrick, L. M., Roffman, J. L., Brady, R. O., Pizzagalli, D. A., Öngür, D., & 778 779 780 Holmes, A. J. (2019). Functional connectomics of affective and psychotic pathology. Proceedings of the National Academy of Sciences, 116(18), 9050–9059. https://doi.org/10.1073/pnas.1820780116 781 Benjamini, Y., & Hochberg, Y. (1995). Controlling the False Discovery Rate: A Practical and Powerful 782 783 Approach to Multiple Testing. Journal of the Royal Statistical Society. Series B (Methodological), 57(1), 289–300. 784 Bijsterbosch, J. D., Woolrich, M. W., Glasser, M. F., Robinson, E. C., Beckmann, C. F., Van Essen, D. C., 785 786 Harrison, S. J., & Smith, S. M. (2018). The relationship between spatial configuration and functional connectivity of brain regions. https://doi.org/10.7554/eLife.32992.001 787 Brown, J. A., Deng, J., Neuhaus, J., Sible, I. J., Sias, A. C., Lee, S. E., Kornak, J., Marx, G. A., Karydas, A. 788 789 790 791 M., Spina, S., Grinberg, L. T., Coppola, G., Geschwind, D. H., Kramer, J. H., Gorno-Tempini, M. L., Miller, B. L., Rosen, H. J., & Seeley, W. W. (2019). Patient-Tailored, Connectivity-Based Forecasts of Spreading Brain Atrophy. Neuron, 104(5), 856-868.e5. https://doi.org/10.1016/J.NEURON.2019.08.037 792 Chong, M., Bhushan, C., Joshi, A. A., Choi, S., Haldar, J. P., Shattuck, D. W., Spreng, R. N., & Leahy, R. 793 794 M. (2017). Individual parcellation of resting fMRI with a group functional connectivity prior. NeuroImage, 156, 87–100. https://doi.org/10.1016/j.neuroimage.2017.04.054 795 Chopra, S., Francey, S. M., O’Donoghue, B., Sabaroedin, K., Arnatkeviciute, A., Cropley, V., Nelson, B., 796 797 798 799 800 Graham, J., Baldwin, L., Tahtalian, S., Yuen, H. P., Allott, K., Alvarez-Jimenez, M., Harrigan, S., Pantelis, C., Wood, S. J., McGorry, P., & Fornito, A. (2021). Functional Connectivity in Antipsychotic-Treated and Antipsychotic-Naive Patients with First-Episode Psychosis and Low Risk of Self-harm or Aggression: A Secondary Analysis of a Randomized Clinical Trial. JAMA Psychiatry, 78(9), 994–1004. https://doi.org/10.1001/jamapsychiatry.2021.1422 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / t / / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t . / f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 801 Dale, A. M., Fischl, B., & Sereno, M. I. (1999). Cortical Surface-Based Analysis: I. Segmentation and 802 803 Surface Reconstruction. NeuroImage, 9(2), 179–194. https://doi.org/10.1006/NIMG.1998.0395 804 Demro, C., Mueller, B. A., Kent, J. S., Burton, P. C., Olman, C. A., Schallmo, M. P., Lim, K. O., & 805 806 Sponheim, S. R. (2021). The psychosis human connectome project: An overview. NeuroImage, 241. https://doi.org/10.1016/j.neuroimage.2021.118439 807 Eickhoff, S. B., Constable, R. T., & Yeo, B. T. T. (2018). Topographic organization of the cerebral 808 809 cortex and brain cartography. NeuroImage, 170, 332–347. https://doi.org/10.1016/j.neuroimage.2017.02.018 810 Eickhoff, S. B., Yeo, B. T. T., & Genon, S. (2018). Imaging-based parcellations of the human brain. 811 Nature Reviews Neuroscience, 19(11), 672–686. https://doi.org/10.1038/s41583-018-0071-7 812 Esteban, O., Birman, D., Schaer, M., Koyejo, O. O., Poldrack, R. A., & Gorgolewski, K. J. (2017). 813 814 MRIQC: Advancing the automatic prediction of image quality in MRI from unseen sites. PLOS ONE, 12(9), e0184661. https://doi.org/10.1371/JOURNAL.PONE.0184661 815 Fischl, B. (2012). FreeSurfer. NeuroImage, 62(2), 774–781. 816 https://doi.org/10.1016/j.neuroimage.2012.01.021 817 Fornito, A., Zalesky, A., & Bullmore, E. T. (2010). Network scaling effects in graph analytic studies of 818 819 human resting-state FMRI data. Frontiers in Systems Neuroscience, 4, 22. https://doi.org/10.3389/fnsys.2010.00022 820 Fornito, A., Zalesky, A., & Bullmore, E. T. (2016). Fundamentals of Brain Network Analysis. Academic 821 Press. https://doi.org/10.1016/B978-0-12-407908-3.09999-4 822 Fornito, A., Zalesky, A., Pantelis, C., & Bullmore, E. T. (2012). Schizophrenia, neuroimaging and 823 824 connectomics. NeuroImage, 62(4), 2296–2314. https://doi.org/10.1016/j.neuroimage.2011.12.090 825 Glasser, M. F., Coalson, T. S., Robinson, E. C., Hacker, C. D., Harwell, J., Yacoub, E., Ugurbil, K., 826 Andersson, J., Beckmann, C. F., Jenkinson, M., Smith, S. M., & Van Essen, D. C. (2016). A l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / / t / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . / t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 827 828 multi-modal parcellation of human cerebral cortex. Nature, 536(7615), 171–178. https://doi.org/10.1038/nature18933 829 Glasser, M. F., Sotiropoulos, S. N., Wilson, J. A., Coalson, T. S., Fischl, B., Andersson, J. L., Xu, J., 830 831 832 Jbabdi, S., Webster, M., Polimeni, J. R., Van Essen, D. C., & Jenkinson, M. (2013). The minimal preprocessing pipelines for the Human Connectome Project. NeuroImage, 80, 105–124. https://doi.org/10.1016/j.neuroimage.2013.04.127 833 Gordon, E. M., Laumann, T. O., Gilmore, A. W., Newbold, D. J., Greene, D. J., Berg, J. J., Ortega, M., 834 835 836 837 Hoyt-Drazen, C., Gratton, C., Sun, H., Hampton, J. M., Coalson, R. S., Nguyen, A. L., McDermott, K. B., Shimony, J. S., Snyder, A. Z., Schlaggar, B. L., Petersen, S. E., Nelson, S. M., & Dosenbach, N. U. F. (2017). Precision Functional Mapping of Individual Human Brains. Neuron, 95(4), 791-807.e7. https://doi.org/10.1016/j.neuron.2017.07.011 838 Goscinski, W. J., McIntosh, P., Felzmann, U., Maksimenko, A., Hall, C., Gureyev, T., Thompson, D., 839 840 841 842 843 Janke, A., Galloway, G., Killeen, N., Raniga, P., Kaluza, O., Ng, A., Poudel, G., Barnes, D., Nguyen, T., Bonnington, P., & Egan, G. (2014). The multi-modal Australian ScienceS Imaging and Visualization Environment (MASSIVE) high performance computing infrastructure: Applications in neuroscience and neuroinformatics research. Frontiers in Neuroinformatics, 8. https://www.frontiersin.org/articles/10.3389/fninf.2014.00030 844 Grabner, G., Janke, A. L., Budge, M. M., Smith, D., Pruessner, J., & Collins, D. L. (2006). Symmetric 845 846 847 848 atlasing and model based segmentation: An application to the hippocampus in older adults. Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention, 9(Pt 2), 58– 66. https://doi.org/10.1007/11866763_8 849 Greve, D. N., & Fischl, B. (2009). Accurate and robust brain image alignment using boundary-based 850 851 registration. NeuroImage, 48(1), 63–72. https://doi.org/10.1016/J.NEUROIMAGE.2009.06.060 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / / t . / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . t / f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 852 Griffanti, L., Salimi-Khorshidi, G., Beckmann, C. F., Auerbach, E. J., Douaud, G., Sexton, C. E., Zsoldos, 853 854 855 856 E., Ebmeier, K. P., Filippini, N., Mackay, C. E., Moeller, S., Xu, J., Yacoub, E., Baselli, G., Ugurbil, K., Miller, K. L., & Smith, S. M. (2014). ICA-based artefact removal and accelerated fMRI acquisition for improved resting state network imaging. NeuroImage, 95, 232–247. https://doi.org/10.1016/J.NEUROIMAGE.2014.03.034 857 HCP Early Psychosis 1.1 Data Release: Reference Manual HUMAN Connectome PROJECT for Early 858 Psychosis. (2021). https://github.com/pnlbwh 859 Hummer, T. A., Yung, M. G., Goñi, J., Conroy, S. K., Francis, M. M., Mehdiyoun, N. F., & Breier, A. 860 861 (2020). Functional network connectivity in early-stage schizophrenia. Schizophrenia Research, 218, 107–115. https://doi.org/10.1016/j.schres.2020.01.023 862 Jenkinson, M., Bannister, P., Brady, M., & Smith, S. (2002). Improved Optimization for the Robust 863 864 and Accurate Linear Registration and Motion Correction of Brain Images. NeuroImage, 17(2), 825–841. https://doi.org/10.1006/NIMG.2002.1132 865 Jenkinson, M., & Smith, S. (2001). A global optimisation method for robust affine registration of 866 867 brain images. Medical Image Analysis, 5(2), 143–156. https://doi.org/10.1016/S1361- 8415(01)00036-6 868 Kong, R., Li, J., Orban, C., Sabuncu, M. R., Liu, H., Schaefer, A., Sun, N., Zuo, X. N., Holmes, A. J., 869 870 871 Eickhoff, S. B., & Yeo, B. T. T. (2019). Spatial Topography of Individual-Specific Cortical Networks Predicts Human Cognition, Personality, and Emotion. Cerebral Cortex, 29(6), 2533– 2551. https://doi.org/10.1093/cercor/bhy123 872 Kong, R., Yang, Q., Gordon, E., Xue, A., Yan, X., Orban, C., Zuo, X.-N., Spreng, N., Ge, T., Holmes, A., 873 874 875 Eickhoff, S., & Yeo, B. T. T. (2021). Individual-Specific Areal-Level Parcellations Improve Functional Connectivity Prediction of Behavior. Cerebral Cortex (New York, N.Y.: 1991), 31(10), 4477–4500. https://doi.org/10.1093/cercor/bhab101 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / / / t . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t . / f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 876 Lewandowski, K. E., Bouix, S., Ongur, D., & Shenton, M. E. (2020). Neuroprogression across the Early 877 878 Course of Psychosis. Journal of Psychiatry and Brain Science, 5, e200002. https://doi.org/10.20900/jpbs.20200002 879 Li, S., Chen, S., Yue, C., & Caffo, B. (2016). A Parcellation Based Nonparametric Algorithm for 880 881 Independent Component Analysis with Application to fMRI Data. Frontiers in Neuroscience, 10. https://www.frontiersin.org/articles/10.3389/fnins.2016.00015 882 Li, T., Wang, Q., Zhang, J., Rolls, E. T., Yang, W., Palaniyappan, L., Zhang, L., Cheng, W., Yao, Y., Liu, Z., 883 884 885 886 Gong, X., Luo, Q., Tang, Y., Crow, T. J., Broome, M. R., Xu, K., Li, C., Wang, J., Liu, Z., … Feng, J. (2017). Brain-Wide Analysis of Functional Connectivity in First-Episode and Chronic Stages of Schizophrenia. Schizophrenia Bulletin, 43(2), 436–448. https://doi.org/10.1093/schbul/sbw099 887 Li, Y., Liu, A., Fu, X., Mckeown, M. J., Wang, Z. J., & Chen, X. (2022). Atlas-guided parcellation: 888 889 Individualized functionally-homogenous parcellation in cerebral cortex. Computers in Biology and Medicine, 150, 106078. https://doi.org/10.1016/j.compbiomed.2022.106078 890 Lynall, M. E., Bassett, D. S., Kerwin, R., McKenna, P. J., Kitzbichler, M., Muller, U., & Bullmore, E. 891 892 (2010). Functional connectivity and brain networks in schizophrenia. Journal of Neuroscience, 30(28), 9477–9487. https://doi.org/10.1523/JNEUROSCI.0333-10.2010 893 McGraw, K. O., & Wong, S. P. (1996). Forming inferences about some intraclass correlation 894 coefficients. Psychological Methods, 1, 30–46. https://doi.org/10.1037/1082-989X.1.1.30 895 Moran, L. V., Tagamets, M. A., Sampath, H., O’Donnell, A., Stein, E. A., Kochunov, P., & Hong, L. E. 896 897 898 (2013). Disruption of anterior insula modulation of large-scale brain networks in schizophrenia. Biological Psychiatry, 74(6), 467–474. https://doi.org/10.1016/j.biopsych.2013.02.029 899 Mueller, S., Wang, D., Fox, M. D., Yeo, B. T. T., Sepulcre, J., Sabuncu, M. R., Shafee, R., Lu, J., & Liu, H. 900 901 (2013). Individual Variability in Functional Connectivity Architecture of the Human Brain. Neuron, 77(3), 586–595. https://doi.org/10.1016/j.neuron.2012.12.028 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / t / / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t / . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 902 Nabulsi, L., McPhilemy, G., Kilmartin, L., Whittaker, J. R., Martyn, F. M., Hallahan, B., McDonald, C., 903 904 905 906 Murphy, K., & Cannon, D. M. (2020). Frontolimbic, Frontoparietal, and Default Mode Involvement in Functional Dysconnectivity in Psychotic Bipolar Disorder. Biological Psychiatry. Cognitive Neuroscience and Neuroimaging, 5(2), 140–151. https://doi.org/10.1016/j.bpsc.2019.10.015 907 Narr, K. L., & Leaver, A. M. (2015). Connectome and schizophrenia. Current Opinion in Psychiatry, 908 28(3), 229–235. https://doi.org/10.1097/YCO.0000000000000157 909 Nogovitsyn, N., Metzak, P. D., Casseb, R. F., Souza, R., Harris, J. K., Prati, L. M., Zamyadi, M., Bray, S. 910 911 912 913 L., Lebel, C., Hassel, S., Strother, S., Goldstein, B. I., Wang, J., Kennedy, S. H., MacQueen, G. M., & Addington, J. (2022). Cerebello-limbic functional connectivity patterns in youth at clinical high risk for psychosis. Schizophrenia Research, 240, 220–227. https://doi.org/10.1016/j.schres.2021.12.041 914 Parkes, L., Fulcher, B., Yücel, M., & Fornito, A. (2018). An evaluation of the efficacy, reliability, and 915 916 sensitivity of motion correction strategies for resting-state functional MRI. NeuroImage, 171, 415–436. https://doi.org/10.1016/J.NEUROIMAGE.2017.12.073 917 Power, J. D., Barnes, K. A., Snyder, A. Z., Schlaggar, B. L., & Petersen, S. E. (2012). Spurious but 918 919 systematic correlations in functional connectivity MRI networks arise from subject motion. Neuroimage, 59(3), 2142. https://doi.org/10.1016/J.NEUROIMAGE.2011.10.018 920 Power, J. D., Lynch, C. J., Silver, B. M., Dubin, M. J., Martin, A., & Jones, R. M. (2019). Distinctions 921 922 among real and apparent respiratory motions in human fMRI data. NeuroImage, 201, 116041. https://doi.org/10.1016/j.neuroimage.2019.116041 923 Power, J. D., Plitt, M., Laumann, T. O., & Martin, A. (2017). Sources and implications of whole-brain 924 925 fMRI signals in humans. NeuroImage, 146, 609–625. https://doi.org/10.1016/j.neuroimage.2016.09.038 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / t . / / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . / t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 926 Rousselet, G. A., Pernet, C. R., & Wilcox, R. R. (2017). Beyond differences in means: Robust graphical 927 928 methods to compare two groups in neuroscience (p. 121079). bioRxiv. https://doi.org/10.1101/121079 929 Salehi, M., Karbasi, A., Shen, X., Scheinost, D., & Constable, R. T. (2018). An exemplar-based 930 931 approach to individualized parcellation reveals the need for sex specific functional networks. NeuroImage, 170, 54–67. https://doi.org/10.1016/j.neuroimage.2017.08.068 932 Salimi-Khorshidi, G., Douaud, G., Beckmann, C. F., Glasser, M. F., Griffanti, L., & Smith, S. M. (2014). 933 934 935 Automatic denoising of functional MRI data: Combining independent component analysis and hierarchical fusion of classifiers. NeuroImage, 90, 449–468. https://doi.org/10.1016/J.NEUROIMAGE.2013.11.046 936 Schaefer, A., Kong, R., Gordon, E. M., Laumann, T. O., Zuo, X.-N., Holmes, A. J., Eickhoff, S. B., & Yeo, 937 938 939 B. T. T. (2018). Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI. Cerebral Cortex, 28(9), 3095–3114. https://doi.org/10.1093/cercor/bhx179 940 Smith, S. M. (2002). Fast robust automated brain extraction. Human Brain Mapping, 17(3), 143–155. 941 https://doi.org/10.1002/HBM.10062 942 Thomas Yeo, B. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M., 943 944 945 946 Roffman, J. L., Smoller, J. W., Zöllei, L., Polimeni, J. R., Fischl, B., Liu, H., & Buckner, R. L. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology, 106(3), 1125–1165. https://doi.org/10.1152/jn.00338.2011 947 Tu, P. C., Lee, Y. C., Chen, Y. S., Li, C. T., & Su, T. P. (2013). Schizophrenia and the brain’s control 948 949 950 network: Aberrant within- and between-network connectivity of the frontoparietal network in schizophrenia. Schizophrenia Research, 147(2–3), 339–347. https://doi.org/10.1016/j.schres.2013.04.011 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / / . / t 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . / t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 951 Welsh, R. C., Chen, A. C., & Taylor, S. F. (2010). Low-frequency BOLD fluctuations demonstrate 952 953 altered thalamocortical connectivity in schizophrenia. Schizophrenia Bulletin, 36(4), 713– 722. https://doi.org/10.1093/schbul/sbn145 954 Whitfield-Gabrieli, S., Thermenos, H. W., Milanovic, S., Tsuang, M. T., Faraone, S. V., Mccarley, R. W., 955 956 957 958 Shenton, M. E., Green, A. I., Nieto-Castanon, A., Laviolette, P., Wojcik, J., Gabrieli, J. D. E., Seidman, L. J., & Raichle, M. E. (2009). Hyperactivity and hyperconnectivity of the default network in schizophrenia and in first-degree relatives of persons with schizophrenia. www.pnas.org/cgi/content/full/ 959 Woodward, N. D., Rogers, B., & Heckers, S. (2011). Functional resting-state networks are 960 961 differentially affected in schizophrenia. Schizophrenia Research, 130(1–3), 86–93. https://doi.org/10.1016/j.schres.2011.03.010 962 Xia, M., Wang, J., & He, Y. (2013). BrainNet Viewer: A Network Visualization Tool for Human Brain 963 Connectomics. PLOS ONE, 8(7), e68910. https://doi.org/10.1371/journal.pone.0068910 964 Yeo, B. T. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M., Roffman, J. L., 965 966 967 Smoller, J. W., Zöllei, L., Polimeni, J. R., Fisch, B., Liu, H., & Buckner, R. L. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology, 106(3), 1125–1165. https://doi.org/10.1152/jn.00338.2011 968 Zalesky, A., Fornito, A., & Bullmore, E. T. (2010). Network-based statistic: Identifying differences in 969 970 brain networks. NeuroImage, 53(4), 1197–1207. https://doi.org/10.1016/J.NEUROIMAGE.2010.06.041 971 Zalesky, A., Fornito, A., Harding, I. H., Cocchi, L., Yücel, M., Pantelis, C., & Bullmore, E. T. (2010). Whole-brain anatomical networks: Does the choice of nodes matter? NeuroImage, 50(3), 970–983. https://doi.org/10.1016/j.neuroimage.2009.12.027 972 973 974 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / . t / / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . t / f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Figure1 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / / t e d u n e n a r t i c e - p d l f / d o i / / / t . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / t . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Figure2 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / . / t / 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d t / . f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Figure4 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / t / e d u n e n a r t i c e - p d l f / d o i / / t / . 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d . / t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Figure3 l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . t / / e d u n e n a r t i c e - p d l f / d o i / . / / t 1 0 1 1 6 2 n e n _ a _ 0 0 3 2 9 2 1 4 2 0 1 4 n e n _ a _ 0 0 3 2 9 p d / . t f b y g u e s t t o n 0 7 S e p e m b e r 2 0 2 3 Levi, P.T., Chopra, S., Pang, J.C., Holmes, A., Gajwani, M., Sassenberg, T.A., DeYoung, C.G. & Fornito, A. (2023). The effect image

PDF Herunterladen