Edge-AI Driven NeuroGaitNet: A Predictive Smart Insole Framework for Early Neuromuscular Disorder Detection and Adaptive Gait Profiling

Abstract

Neuromuscular disorders (NMDs) such as Parkinson, Amyotrophic Lateral Sclerosis (ALS) and Peripheral Neuropathy are the progressive disorders that severely affect motor interaction and living standards. Early detection is vital for early intervention but conventional means and methods of diagnosis miss the finer points of gait-related anomalies in the prodromal periods. This article proposes an intelligent gait-based framework (NeuroGaitNet) to predict NMD within a population at an early stage. This system uses MultiZone Adaptive Pressure Mapping (MZAPM) for the measurement of local plantar pressures and SpatioTemporal Harmonic Encoding (STHE) for joint spatial temporal gait feature extraction. Individualized baselines are made and the deviations are qualified with Neuro-Signature Deviation Index (NSDI), so that the nuances of abnormality are identified in relation to body-specific patterns. Privacy-preserving Edge Adaptive Federated Learning (EA-FL) is the concept of distributed training of a model without sharing raw data. Hence, the adaptation is ensured at the patient-specific level and it is not violated. A Predictive Risk Stratification Layer (PRSL) integrates smoothed NSDI scores, calibrated probabilities and stability measures towards the capability of creating the clinically meaningful risk levels. Experimental analysis contrasted NeuroGaitNet from the existing models in terms of accuracy, precision, recall, F-measure as well as specificity. Experimental results show that the proposed MZAPM model performed better, exhibiting better performance in all the metrics. In particular, it has achieved maximum accuracy (92.4%), precision, recall and specificity which confirm its strength in the differentiation of pathological gait and natural variability. The fact that the precision and recall rose indicated that the algorithm is capable of reducing false-positives (37%) and identifying the genuine cases of anomalies. Local personalization and global knowledge hybrid treatment scheme of learning increased the sensitivity of detection. In general, NeuroGaitNet offers a safe, privacy-preserving and scalable architecture that is further used in the early detection of NMD based on the intelligent gait analysis.