A novel deep dual self-attention and Bi-LSTM fusion framework for Parkinson's disease prediction using freezing of gait: a biometric application
Zeeshan Habib,Muhammad Ali Mughal,Muhammad Attique Khan,Ameer Hamza,Nazik Alturki,Leila Jamel
DOI: https://doi.org/10.1007/s11042-024-18906-5
IF: 2.577
2024-03-22
Multimedia Tools and Applications
Abstract:Parkinson's disease (PD) disorder is caused by the imbalance of inhibitory dopamine and excitatory acetylcholine neurotransmitters, which causes hindrance in locomotion. Freezing of gait (FOG), tremors, and bradykinesia are the most debilitating and disconcerting symptoms of Parkinson's Disease (PD). FOG episodes occur due to neurological control disorder and motorized impairments, which sternly hinder voluntary movement and forward locomotion. The computational intelligence techniques can effectively predict PD from the sensor data. To detect the FOG episodes often experienced by PD patients, this paper utilizes a 5G spectrum that operates at 4.8 GHz. The main idea of the proposed system is to extract the wireless channel characteristics information (containing amplitude variances) that can be integrated into a 5G communication system. This system collects data through wireless devices such as dipole antennas, network interface cards, and RF signal generators. Five human activities were performed: fast walking, slow walking, sitting on a chair, FOG episodes, and voluntary stopping. We proposed a novel deep dual self-attention and Bi-LSTM fusion architecture for PD prediction to train the collected dataset. Two architectures have been proposed, namely Deep Dual Attention Neural Network (D2ANN) and BiLSTM Neural Network (BNN), for the classification of PD using Chanel State Information (CSI) data. After that, the most prominent information is extracted using self-attention activation. A newly informed weighted fusion approach is proposed, and tree growth optimization is employed for the best feature selection. Finally, the selected features are classified using a multilayer full SoftMax neural network. The experimental process was conducted on FOG wireless sensor network data and obtained the highest accuracy of 98.66%. Compared to the results with state-of-the-art techniques, the proposed framework shows improved accuracy, recall, and precision rates.
computer science, information systems, theory & methods,engineering, electrical & electronic, software engineering