Search for: All records

We present a novel deep learning framework to automatically compute independently salient networks in the brain that characterize the underlying changes in the brain in association with clinically observed assessments. Unsupervised approaches for high-dimensional neuroimaging data focus on computing low-dimensional brain components for subsequent analysis, while supervised learning approaches aim for predictive performance and yielding a single list of associative feature importance, thus making it hard to interpret at the level of brain subsystems. Our approach integrates the goals of decomposition into lower dimensional subspaces and, identifying salient brain subsystems into a single automated framework. We first train a convolutional neural network on structural brain features to predict clinical assessments, followed by a multi-step decomposition in the saliency space to compute salient brain networks that intrinsically characterize the brain changes associated with the assessment. Through a repeated training procedure on an Alzheimer’s disease (AD) dataset, we show that our method effectively computes AD-related salient brain subsystems directly from high-dimensional neuroimaging data, while maintaining predictive performance. Such approaches are crucial for data-driven biomarker development for brain disorders. 

Neurological disorders generally involve multiple kinds of changes in the functional and structural properties of the brain. In this study, we develop a CNN-based multimodal deep learning pipeline by exploiting both functional and structural neuroimaging features to generate full-brain maps that encode significant differences between patient groups and between modalities in terms of their distinctive contribution towards diagnostic classification of Alzheimer’s disease. Through a repeated cross-validation procedure and robust statistical analysis, we show that our approach can be used to encode highly discriminative and abstract information from full-brain data, while also retaining the ability to identify and categorize significantly contributing voxel-level features based on their salient strength in various diagnostic and modality-related contexts. Our results on an Alzheimer’s disease classification task show that such approaches can be used for creating more elaborately defined biomarkers for brain disorders.