Population heterogeneity in clinical cohorts affects the predictive accuracy of brain imaging

• DOI: 10.1371/journal.pbio.3001627
• Publication Date: 2022-04-29T18:02:57Z
• Authors: Oualid Benkarim; Casey Paquola; Bo-yong Park; Valeria Kebets; Seok-Jun Hong; Reinder Vos de Wael; Shaoshi Zhang; B. T. Thomas Yeo; Michael Eickenberg; Tian Ge; Jean-Baptiste Poline; Boris C. Bernhardt; Danilo Bzdok
• Keywords: QH301-705.5; Brain; Neuroimaging; Magnetic Resonance Imaging; 3. Good health; 03 medical and health sciences; 0302 clinical medicine; Humans; Supervised Machine Learning; Biology (General); Algorithms; info:eu-repo/classification/ddc/610; Research Article

Brain imaging research enjoys increasing adoption of supervised machine learning for single-participant disease classification. Yet, the success these algorithms likely depends on population diversity, including demographic differences and other factors that may be outside primary scientific interest. Here, we capitalize propensity scores as a composite confound index to quantify diversity due major sources variation. We delineate impact heterogeneity predictive accuracy pattern stability in 2 separate clinical cohorts: Autism Imaging Data Exchange (ABIDE, n = 297) Healthy Network (HBN, 551). Across various analysis scenarios, our results uncover extent which cross-validated prediction performances are interlocked with diversity. The instability extracted brain patterns attributable is located preferentially regions part default mode network. Collectively, findings highlight limitations prevailing deconfounding practices mitigating full consequences