Early Diagnosis of Alzheimer’s Disease using Multiscale and Multipath Convolutional Neural Network

Early diagnosis of Alzheimer’s disease is inevitably essential to prevent the progression of the disease to its final stage. It helps the patient to be aware of the severity of the disease and can take necessary therapeutic and medication to prevent the progression. An MRI is a good biomarker to detect the disease at an early stage, known as Mild Cognitive Impairment(MCI). This work aims at performing a ternary classification using 2D MRI images to simultaneously diagnose MCI, AD and Normal in a single algorithm. Due to the subtle nature of the atrophy regions, discriminating between the classes is cumbersome, so here we propose a multiscale and multipath convolution neural network to learn discriminating features for proper classification among the three categories. The shallow and deep layers features are concatenated using multipath mode for better feature learning and avoiding vanishing gradient problems. In multi-scale, low-rank convolutional kernels in parallel extract multiscale features, which enhances the feature extraction property of the network.The proposed model is trained and tested on ADNI dataset comprising 900 subjects to give an accuracy of 95.19% with tenfold cross-validation. The multiscale and multipath method significantly reduces the number of learnable parameters