Ten simple rules for predictive modeling of individual differences in neuroimaging

Scheinost, Dustin; Noble, Stephanie; Horien, Corey; Greene, Abigail S.; Lake, Evelyn; Salehi, Mehraveh; Gao, Siyuan; Shen, Xilin; O’Connor, David; Barron, Daniel S.; Yip, Sarah W.; Rosenberg, Monica D.; Constable, R. Todd

doi:10.1016/j.neuroimage.2019.02.057

Cited by 312 publications

(333 citation statements)

References 73 publications

(111 reference statements)

Supporting

Mentioning

302

Contrasting

Order By: Relevance

“…For a predictive model to be useful in real-world applications, it needs to generalize well to datasets from different sites (Scheinost et al, 2019). While characteristics of our study facilitate generalization, a future study is required to empirically establish the generalization of our models to independent datasets.…”

Section: Discussionmentioning

confidence: 99%

Predicting future cognitive decline from non-brain and multimodal brain imaging data in healthy and pathological aging

Liem

Dadi

Engemann

et al. 2020

Preprint

View full text Add to dashboard Cite

Cognitive decline occurs in healthy and pathological aging, and both may be preceded by subtle changes in the brain — offering a basis for cognitive predictions. Previous work has largely focused on predicting a diagnostic label from structural brain imaging. Our study broadens the scope of applications to cognitive decline in healthy aging by predicting future decline as a continuous trajectory, rather than a diagnostic label. Furthermore, since brain structure as well as function changes in aging, it is reasonable to expect predictive gains when using multiple brain imaging modalities. Here, we tested whether baseline multimodal neuroimaging data improve the prediction of future cognitive decline in healthy and pathological aging. Non-brain data (including demographics and clinical and neuropsychological scores) were combined with structural and functional connectivity MRI data from the OASIS-3 project (N = 662; age = 46 – 96y). The combined input data was entered into cross-validated multi-target random forest models to predict future cognitive decline (measured by the Clinical Dementia Rating and the Mini-Mental State Examination), on average 5.8y into the future. The analysis was preregistered and all analysis code is publicly available. We found that combining non-brain with structural data improved the continuous prediction of future cognitive decline (best test-set performance: R2 = 0.42) and that cognitive performance, daily functioning, and subcortical volume drove the performance of our model. In contrast, including functional connectivity did not improve predictive accuracy. In the future, the prognosis of age-related cognitive decline may enable earlier and more effective cognitive, pharmacological, and behavioral interventions to be tailored to the individual.

show abstract

Section: Discussionmentioning

confidence: 99%

Predicting future cognitive decline from non-brain and multimodal brain imaging data in healthy and pathological aging

Liem

Dadi

Engemann

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…That is, in evaluating and weighting each component of the signal separately, the combined model is able to capture more information than is contained in any of its component parts, or in comparably preprocessed standard FC (Figure 2—figure supplement 3), even when incorrect task regressors are used (Figure 2—figure supplement 4). Similar efforts to reveal brain-phenotype relationships may therefore benefit from the inclusion of more features—even if their relationship to the phenotype of interest is relatively weak—in models that use regularization (Gao et al, 2019), although it is always wise to exclude uninformative features to avoid overfitting (Scheinost et al, 2019).…”

Section: Discussionmentioning

confidence: 99%

“…A modified version of connectome-based predictive modeling (CPM; Finn et al, 2015; Shen et al, 2017) was used to predict gF from brain measures (i.e., beta matrices [see Psychophysiological interaction analysis ]) using ridge regression (Gao et al, 2019). This pipeline predicts gF in novel subjects, validating the model through iterative, k -fold cross-validation; in this work, k = 10 to balance model bias and variance given the large sample size (Scheinost et al, 2019). Consistent with this motivation, split-half (i.e., k = 2) analyses yielded comparable patterns of results (e.g., best performance from combined models), but overall lower prediction performance (Figure 2—figure supplement 8).…”

Section: Methodsmentioning

confidence: 99%

How tasks change whole-brain functional organization to reveal brain-phenotype relationships

Greene

Gao

Noble

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

Functional connectivity (FC) calculated from task fMRI data better reveals brain-phenotype relationships than rest-based FC, but why is unknown. In over 700 individuals performing 7 tasks, we use psychophysiological interaction (PPI) and predictive modeling analyses to demonstrate that FC and overall degree of task-induced signal change, but not task-evoked activation alone, drive phenotypic prediction, and their combination further improves prediction. Inter-subject PPI demonstrates that predictive utility is highest in distributed FC patterns that are dissimilar across individuals, except in regions of group-level task activation, suggesting that task FC better predicts phenotype than rest FC for two, regionally specific reasons: (1) tasks synchronize activated regions and amplify signal components that meaningfully vary across individuals; and (2) elsewhere, prediction is driven by nodal interactions that set individuals apart. These findings offer a framework to leverage both task activation and FC to reveal the neural bases of complex human traits, symptoms, and behaviors.

show abstract

“…Based on the whole-brain rsFC, we used CPM to predict the degree of anxiety individually. In light of ten simple rules for applying predictive modeling to rsFC data (sample size < 200; Scheinost et al, 2019) and to be consistent with past work employing CPM Rosenberg et al, 2015;Beaty et al, 2018), we performed leave-one-out cross-validation (LOOCV). That is, in each iterative analysis, the predictive model was built based on n -1 participants (training set) and then the score of the remaining participant (test set) was predicted.…”

Section: Connectome-based Predictive Modeling (Cpm)mentioning

confidence: 99%

Connectome-based predictive modeling of individual anxiety

Wang

Goerlich

et al. 2020

Preprint

View full text Add to dashboard Cite

Anxiety-related illnesses are highly prevalent in human society. Being able to identify neurobiological markers signaling high trait anxiety could aid the assessment of individuals with high risk for mental illness. Here, we applied connectome-based predictive modeling (CPM) to whole-brain resting-state functional connectivity (rsFC) data to predict the degree of anxiety in 76 healthy participants. Using a computational "lesion" method in CPM, we then examined the weights of the identified main brain areas as well as their connectivity. Results showed that the CPM could predict individual anxiety from whole-brain rsFC, especially from limbic areas-whole brain and prefrontal cortex-whole brain. The prediction power of the model significantly decreased from (simulated) lesions of limbic areas, lesions of the connectivity within the limbic system, and lesions of the connectivity between limbic regions and the prefrontal cortex.Although the same model also predicted depression, anxiety-specific networks could be identified independently, centered at the prefrontal cortex. These findings highlight the important role of the limbic system and the prefrontal cortex in the prediction of anxiety.Our work provides evidence for the usefulness of connectome-based modeling of rsFC in predicting individual personality differences and indicates its potential for identifying personality structures at risk of developing psychopathology.

show abstract

Ten simple rules for predictive modeling of individual differences in neuroimaging

Cited by 312 publications

References 73 publications

Predicting future cognitive decline from non-brain and multimodal brain imaging data in healthy and pathological aging

Predicting future cognitive decline from non-brain and multimodal brain imaging data in healthy and pathological aging

How tasks change whole-brain functional organization to reveal brain-phenotype relationships

Connectome-based predictive modeling of individual anxiety

Contact Info

Product

Resources

About