Parkinson's illness Deep Learning Diagnosis: An Innovative LSTM-Based Method for Freezing Gait Detection

By uncovering hidden patterns in big clinical datasets, deep learning has great promise for the medical industry in terms of aiding in the diagnosis of a wide range of diseases. A deterioration in brain function is a hallmark of Parkinson's disease (PD), a neurodegenerative condition. Early automated detection of Parkinson's disease is challenging due to the behavioral similarities between those with the disease and healthy individuals. Our objective is to offer a practical model that can facilitate the early detection of Parkinson's disease. We utilized the VGRF gait signal dataset, which was acquired via Physionet, to distinguish between individuals with Parkinson's disease and healthy individuals. A novel deep learning architecture based on LSTM networks is presented in this study to automatically detect freezing of gait episodes in Parkinson's disease. Unlike typical machine learning techniques, this method effectively captures long-term temporal correlations in gait patterns and eliminates the requirement for human feature engineering, improving the diagnosis of Parkinson's disease. To avoid the issue of vanishing gradients and enable optimal information absorption, the LSTM network uses memory blocks instead of selfconnected hidden units. Methods such as L2 regularization and dropout have been employed to prevent overfitting. Adam, an optimizer based on stochastic gradients, is also used in the optimization process. The results demonstrate that our proposed approach, with 97.71% accuracy, 99% sensitivity, 98% precision, and 96% specificity, surpasses the state-of-the-art models in FOG episode recognition. This demonstrates how promising it is as an improved classification method for Parkinson's disease diagnosis.