Machine learning techniques have become increasingly popular in the field ofresting state fMRI (functional magnetic resonance imaging) network basedclassification. However, the application of convolutional networks has beenproposed only very recently and has remained largely unexplored. In this paperwe describe a convolutional neural network architecture for functionalconnectome classification called connectome-convolutional neural network(CCNN). Our results on simulated datasets and a publicly available dataset foramnestic mild cognitive impairment classification demonstrate that our CCNNmodel can efficiently distinguish between subject groups. We also show that theconnectome-convolutional network is capable to combine information from diversefunctional connectivity metrics and that models using a combination ofdifferent connectivity descriptors are able to outperform classifiers usingonly one metric. From this flexibility follows that our proposed CCNN model canbe easily adapted to a wide range of connectome based classification orregression tasks, by varying which connectivity descriptor combinations areused to train the network.
展开▼
