Parkinson's Disease Severity Assessment Based on Hybrid Features and Optimized Stacked Long Short-Term Memory

Parkinson’s disease (PD) is a common neurological condition that primarily affects middle-aged and older people, making it difficult for them to move around and be flexible. Slow motion, tremors in the hands, feet, or other body parts, and stiffness and loss of flexibility are the main features of Parkinson's disease. Thus, the use of computer-aided methods in PD diagnosis is essential. This work presents a novel framework for efficiently extracting local and global features from gait images. This article presents a novel approach that uses hybrid features and an optimized stacked long short-term memory (Opt-StLSTM) network to evaluate the severity of Parkinson's disease. The Enhanced DenseCapsNet (EDCN) model, which combines multi-level capsules with densely connected convolutions, is utilized to capture global features. At the same time, a Variational Autoencoder (VAE) is employed to extract local features. The Opt-StLSTM network, which has been optimized by the Adaptive Red Panda Optimization (ARPO) method, is then used to classify the fused features. The efficacy of the proposed technique is demonstrated with the GAIT-IT and GAIT-IST datasets. Accuracy, precision, recall, and F-measure are used to evaluate its performance in comparison to cutting-edge techniques.