Alzheimer's Disease Brain Network Classification Using Improved Transfer Feature Learning with Joint Distribution Adaptation

Alzheimer's disease significantly affects the quality of life of patients. This paper proposes an approach to identify Alzheimer's disease based on transfer learning using functional MRI images, which is especially useful when the training dataset is small. Transfer learning improves the performance of the classifier with the help of an auxiliary dataset, which may be obtained from a different population group and/or machine. First, we used the joint distribution adaptation method to project the source and target domain samples into a new feature space, and then we built a classifier that works well in both the source and target domains but emphasizes the target domain. In the classifier, we assigned larger weights to the target domain samples and minimized the weighted loss in classifying the samples in both domains. Experimental results verify the effectiveness of our proposed approach and, with the help of the auxiliary samples, the classification accuracy of our target dataset has been greatly improved.