Classification of Brain MRI Slices into Control/Alzheimer's using Lightweight Deep-Learning with Fused Features
V. Rajinikanth
Abstract
The disease occurrence rate in elderly individual is more due to various causes. The Alzheimer ’s disease (AD) in aged individual will causes damage to memory as well as behaviour. The clinical level detection of AD is commonly performed using the image supported techniques and Magnetic Resonance Imaging (MRI) is one of the commonly considered approaches. Appropriate detection and handling of the AD is necessary, hence computer algorithm supported AD detection is widely considered. This work proposes a lightweight deep-learning scheme (LDLS) to detect the AD in axial-plane MRI-database. The stages of this LDLS includes; MRI slice collection and resizing it to 224x224 pixels, feature extraction using deep-learning models, identification of best model based on the accuracy achieved with SoftMax classifier, implementing 50% dropout on selected models to reduce the features, and executing a serial features concatenation to achieve fused-feature vector (FV), and verifying the performance using 3-fold cross-validation. This work considered 1000 images per class and the implemented scheme substantiate merit of FV based on the achieved accuracy (>98% ), when random-forest classifier is considered.