Multi-site benchmark classification of major depressive disorder using machine learning on cortical and subcortical measures
Vladimir Belov, Tracy Erwin-Grabner, Ali Saffet Gonul, Alyssa R. Amod, Amar Ojha, Andre Aleman, Annemiek Dols, Anouk Scharntee, Aslihan Uyar-Demir, Ben J Harrison, Benson M. Irungu, Bianca Besteher, Bonnie Klimes-Dougan, Brenda W. J. H. Penninx, Bryon A. Mueller, Carlos Zarate, Christopher G. Davey, Christopher R. K. Ching, Colm G. Connolly, Cynthia H. Y. Fu, Dan J. Stein, Danai Dima, David E. J. Linden, David M. A. Mehler, Edith Pomarol-Clotet, Elena Pozzi, Elisa Melloni, Francesco Benedetti, Frank P. MacMaster, Hans J. Grabe, Henry Völzke, Ian H. Gotlib, Jair C. Soares, Jennifer W. Evans, Kang Sim, Katharina Wittfeld, Kathryn Cullen, Liesbeth Reneman, Mardien L. Oudega, Margaret J. Wright, Maria J. Portella, Matthew D. Sacchet, Meng Li, Moji Aghajani, Mon-Ju Wu, Natalia Jaworska, Neda Jahanshad, Nic J. A. van der Wee, Nynke Groenewold, Paul J. Hamilton, Philipp Saemann, Robin Bülow, Sara Poletti, Sarah Whittle, Sophia I. Thomopoulos, Steven J. A. van, der Werff, Sheri-Michelle Koopowitz, Thomas Lancaster, Tiffany C. Ho, Tony T. Yang, Zeynep Basgoze, Dick J. Veltman, Lianne Schmaal, Paul M. Thompson, Roberto Goya-Maldonado
Unverified — Be the first to reproduce this paper.
ReproduceAbstract
Machine learning (ML) techniques have gained popularity in the neuroimaging field due to their potential for classifying neuropsychiatric disorders. However, the diagnostic predictive power of the existing algorithms has been limited by small sample sizes, lack of representativeness, data leakage, and/or overfitting. Here, we overcome these limitations with the largest multi-site sample size to date (n=5,356) to provide a generalizable ML classification benchmark of major depressive disorder (MDD). Using brain measures from standardized ENIGMA analysis pipelines in FreeSurfer, we were able to classify MDD vs healthy controls (HC) with around 62% balanced accuracy, but when harmonizing the data using ComBat balanced accuracy dropped to approximately 52%. Similar results were observed in stratified groups according to age of onset, antidepressant use, number of episodes and sex. Future studies incorporating higher dimensional brain imaging/phenotype features, and/or using more advanced machine and deep learning methods may achieve more encouraging prospects.