Multimodal Deep Learning for Freezing of GaitDetection in Parkinson’s Disease UsingWearable Sensors

Mosleh Al-Adhaileh¹Asim Wadood²Theyazn H.H Aldhyani^1*Safeer Khan²Irfan Uddin²Abdullah H Al-Nefaie¹

• ¹King Faisal University, Al-Ahsa, Saudi Arabia
• ²Kohat University of Science and Technology, Kohat, Khyber Pakhtunkhwa, Pakistan

Freezing of Gait (FoG) is a disabling symptom in Parkinson’s Disease (PD) that substantially4 affects patient ability to movement and hence quality of life. The current research aims to introduce5 a framework that is based on advanced AI techniques such as deep learning techniques that6 utilizes a more special type of deep learning models known as Convolutional Neural Networks7 (CNNs) for spatial characteristics discover and a sequence based model known as Bidirectional8 Long Short-Term Memory (BiLSTM) networks for temporal modeling, complemented by an9 attention mechanism to pinpoint essential gait features. The approach is trained with multimodal10 sensor data across sources like tDCS FoG, DeFOG, Daily Living, and Hantao’s Multimodal11 datasets in order to establish robustness and generalizability over populations of varied natures.12 Preeminent issues of sensor noise, inter-subject variation, and class imbalance are countered13 with deep preprocessing operations, i.e., sensor fusion, normalization, and data augmentation.14 The model exhibits high performance with accuracy (98%), F1-score (94%), and AUC (96%) over15 existing methods for detecting FoG episodes. The present research shows the ability of AI-aided16 analysis of wearable sensor data in enhancing diagnostic accuracy, therapy personalization, and17 monitoring of unfortunate people having neurological disability and advances the wider context of18 AI application in medicine.