Alzheimer's Disease detection and classification using optimized neural network

Abstract

Alzheimer's disease (AD) is a degenerative neurological condition characterized by a progressive decline in cognitive abilities, resulting in memory impairment and limitations in performing daily tasks. Timely and precise identification of AD holds paramount importance for prompt intervention and enhanced patient prognosis. In this research, a novel approach to AD mechanism was developed by combining Deep Reinforcement Learning (DRL) with a Moth Flame Optimized Recurrent Neural Network (MFORNN). Initially, the brain MRI samples are gathered and preprocessed to discard the noise features and to improve their quality. Consequently, the MFO algorithm captures and selects the most informative and highly correlative features from the preprocessed images, making it easier for Recurrent Neural Networks (RNNs) to learn the temporal dependencies and patterns differentiating normal and AD-affected images. The DRL component fine-tunes the parameters of RNN through its reward-based mechanism, ensuring that the classifier produces accurate outcomes and reduces computational complexity. The Python tool was utilized to implement the outlined framework, with the outcomes showcased that the designed algorithm attained an accuracy of 99.31 %, precision of 99.24 %, recall of 99.43 %, and f-measure of 99.35 %. Ultimately, a comparative analysis was performed against established classifier models, affirming the superior performance of the proposed technique over conventional algorithms.