Parkinson’s disease (PD) is a progressive neurodegenerative disorder characterized by motor impairments such as tremor, rigidity, bradykinesia, and gait instability. Accurate and early detection remains challenging due to the complex temporal nature of wearable sensor signals, limited interpretability of deep learning models, and concerns regarding medical data integrity. This study proposes a novel Blockchain-Enabled Hybrid CNN–Transformer framework with Integrated Gradients for PD detection using smartwatch-based inertial sensor data. The CNN module extracts local motion features, while the Transformer captures long-range temporal dependencies within movement signals. To enhance transparency, Integrated Gradients is employed to identify the most influential signal regions contributing to model predictions. In addition, a lightweight blockchain layer is incorporated to provide tamper-resistant storage and traceability of diagnostic outcomes. Experimental evaluation on the PADS dataset achieved 97.29% accuracy, 97.14% precision, 97.29% recall, 97.12% F1-score, and an AUC of 0.9869. The results demonstrate a reliable, interpretable, and secure framework for wearable-based Parkinson’s disease detection.
Jigna J. Hathaliya, Hetav Modi, Rajesh Gupta, Sudeep Tanwar
Essential tremor (ET) and Parkinson’s tremor (PST) are neurological movement disorders in which ET emerges with body part activation, while PST is recorded in the relaxed position of the patient. The medical symptoms of ET and PST are equivalent, including gait, anxiety, and muscular stiffness. In both disorders, doctors diagnose patients related to clinical evaluations during such hospital visits, leading to misdiagnosis. Machine Learning (ML) is being used to classify the ET and PST using human-based feature extraction to address this issue. Motivated by this, we applied Deep Learning (DL) to overcome the ML issue via automating feature extraction through the model itself. In this paper, we have used the integration of Gated recurrent unit (GRU) and Long short term memory (LSTM) algorithms to predict tremor severity. Initially, accelerometer sensors are used to record tremors in all three axial dimensions for each subject. Further, this data is pre-processed using the standard scalar function and scaled in-unit variance. Furthermore, this data first passed through the GRU model, and later it fed into the LSTM model to improve the model’s performance. Moreover, we employed the blockchain (BC) network to validate the performance of the trained model. we have used a smart contract to validate the identity of the researcher. The proposed model outperforms with 80.4% training accuracy and 74.1% testing accuracy. The total communication and computation cost of the proposed scheme is 448 bits and 0.056 ms. The integration of BC and DL makes a system more reliable, transparent, and accurate.