Develop a novel, faster mask region-based convolutional neural network model with leave-one-subject-out to predict freezing of gait abnormalities of Parkinson’s disease

• Published in: Neural computing & applications Vol. 37; no. 7; pp. 5441 - 5457
• Main Authors: Ezhilarasi, J., Senthil Kumar, T.
• Format: Journal Article
• Language: English
• Published: London Springer London 01.03.2025; Springer Nature B.V
• Subjects: Abnormalities; Accelerometers; Algorithms; Artificial Intelligence; Artificial neural networks; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Deep learning; Gait; Image Processing and Computer Vision; Machine learning; Neural networks; Original Article; Parkinson's disease; Probability and Statistics in Computer Science; Signal processing; Deep learning; Freezing of gait; Signal processing; Faster mask recurrent-based convolutional neural network; Parkinson’s disease; Long short term memory
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-024-10832-9

A common symptom of severe Parkinson’s disease (PD) is Freezing of Gait (FoG), a gait disorder that causes sudden difficulty in initiating or maintaining walking. FoG frequently leads to falls and has a detrimental impact on a patient’s regular life. Real-time detection algorithms identify FoG occurrences using wearable sensors. Anticipating FoG in advance allows for pre-emptive cueing, which may prevent the episodes or reduce their severity and duration. This research proposes a Faster Mask Region-based Convolutional Neural Network (FMRCNN) signal processing approach for FoG identification. The model captured gyroscope, magnetometer, and tri-axial accelerometer signals using an inertial measurement device on the left side of the abdomen. The experimental results demonstrate a reduction in the equal error rate to 1.9% in the Leave-One-Subject-Out (LOSO) Cross-Validation (CV) with Long Short Term Memory (LSTM) assessment. Additionally, the tenfold CV evaluation enhances specificity and sensitivity by 0.045 and 0.017, respectively, compared to previous best results. It takes only 0.52 ms to detect a 256-data section. The proposed work uses the LOSO-CV-LSTM to evaluate various machine learning (ML) and deep learning (DL) techniques for FoG detection. The proposed system not only detects FoG but also enhances the automation of PD detection and therapy at an earlier stage. The results demonstrate that the proposed system improves performance measures compared to existing systems.