Enhanced real-time Parkinson’s disease monitoring and severity prediction using a multi-faceted deep learning approach

Parkinson’s disease (PD) is an escalating neurological disorder that is primarily characterized by motor symptoms such as tremors, rigidity, bradykinesia, and balance impairment. Tremors, an early symptom of PD, manifest as rhythmic shaking in limbs or jaw. Current diagnostic methods struggle with capturing complex temporal and spatial patterns, lack in real-time analysis and scalability for wearable devices. This research proposes a novel deep learning framework using a convolutional long short-term memory (LSTM) network to detect tremor anomalies in PD patients. The model was trained and validated across two datasets containing more than 1000 patient records with sensor-derived tremor measurements. The proposed robust architecture incorporates feature extraction using convolutional layers, and a spatial dropout mechanism for reducing overfitting. This helps the model learn robust features that are invariant to specific paths within the network and LSTM layers to capture temporal dependencies. The proposed model achieved 99.62% accuracy, mean squared error (MSE) of 0.44 and mean absolute error (MAE) of 0.45. The proposed model R-squared (R²) value of 0.996 indicates its potential for early diagnosis and continuous PD management by monitoring tremor severity and treatment response. The Parkinson’s Assessment and Detection Model (PADM) enhances diagnostic precision in real-time personalized patient care.