Applicability of Manually Crafted Convolutional Neural Network for Classification of Mild Cognitive Impairment

Abstract

Mild Cognitive Impairment (MCI) is considered as a formative stage of dementia and therefore its diagnosis can significantly assist in providing apposite treatment to the patients to impediment its headway towards dementia. In this paper, a Deep Learning approach is proposed for the classification of MCI-Converts and MCI-Non Converts, using the Structural Magnetic Resonance Imaging data. It investigates the effect of the variation in the number of filters, and the size of the filter on the performance of the model. Furthermore, the features are extracted using the penultimate layer of the proposed architecture. The Fisher Discriminant Ratio is used for the selection of features and the Support Vector Machine for the classification. The results are also compared to those obtained using the Softmax Layer. The proposed pipeline is able to extort germane features, thus improving the classification accuracy. The empirical studies exhibit the supremacy of the proposed method over the existing ones, in terms of accuracy. Consequently, the proposed technique may prove useful in the effectual diagnosis of MCI.