Ensemble learning with 3D convolutional neural networks for functional connectome-based prediction

• Published in: NeuroImage (Orlando, Fla.) Vol. 199; pp. 651 - 662
• Main Authors: Khosla, Meenakshi, Jamison, Keith, Kuceyeski, Amy, Sabuncu, Mert R.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.10.2019; Elsevier Limited
• Subjects: ABIDE; Accuracy; Adolescent; Adult; Age; Artificial intelligence; Atlases as Topic; Autism; Autism spectrum disorder; Autism Spectrum Disorder - diagnostic imaging; Brain - diagnostic imaging; Brain - physiopathology; Brain mapping; Child; Classification; Cohort Studies; Connectome - methods; Connectome - standards; Convolutional neural networks; Datasets; Deep learning; Dictionaries; fMRI; Functional connectivity; Functional magnetic resonance imaging; Humans; Image Interpretation, Computer-Assisted - methods; Image Interpretation, Computer-Assisted - standards; Learning algorithms; Machine Learning; Magnetic Resonance Imaging - methods; Magnetic Resonance Imaging - standards; Male; Neural networks; Neural Networks, Computer; Prediction models; Statistical analysis; Young Adult; Functional connectivity; fMRI; ABIDE; Convolutional neural networks; Autism spectrum disorder
• ISSN: 1053-8119; 1095-9572; 1095-9572
• DOI: 10.1016/j.neuroimage.2019.06.012

The specificity and sensitivity of resting state functional MRI (rs-fMRI) measurements depend on preprocessing choices, such as the parcellation scheme used to define regions of interest (ROIs). In this study, we critically evaluate the effect of brain parcellations on machine learning models applied to rs-fMRI data. Our experiments reveal an intriguing trend: On average, models with stochastic parcellations consistently perform as well as models with widely used atlases at the same spatial scale. We thus propose an ensemble learning strategy to combine the predictions from models trained on connectivity data extracted using different (e.g., stochastic) parcellations. We further present an implementation of our ensemble learning strategy with a novel 3D Convolutional Neural Network (CNN) approach. The proposed CNN approach takes advantage of the full-resolution 3D spatial structure of rs-fMRI data and fits non-linear predictive models. Our ensemble CNN framework overcomes the limitations of traditional machine learning models for connectomes that often rely on region-based summary statistics and/or linear models. We showcase our approach on a classification (autism patients versus healthy controls) and a regression problem (prediction of subject's age), and report promising results. [Display omitted] •We propose 3D CNN framework for functional connectomes.•Ensemble learning with stochastic parcellations outperform atlas-based models.•We showcase our method for classification of autism vs. healthy and predicting age.