We proposed a novel multi-filter convolutional neural network for prediction of cognitive deficits using brain structural connectome data. In contrast to 2D grid convolutional filters in traditional convolutional neural networks, our proposed model contains multiple vector-shape convolutional filters that can better extract the topological information from brain connectome. We demonstrated the ability of our model to learn hidden patterns from brain connectome data for prediction tasks. Our proposed model was able to identify infants at a high risk of cognitive deficits with an area under the curve of 0.78, exceeding the performance of other existing peer convolutional neural network methods.
This abstract and the presentation materials are available to members only; a login is required.