Advanced multiscale machine learning for nerve conduction velocity analysis

• Published in: Scientific reports Vol. 15; no. 1; pp. 23399 - 13
• Main Author: Sadeghi, Hossein
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 02.07.2025; Nature Publishing Group; Nature Portfolio
• Subjects: 631/378; 631/378/3920; Accuracy; Deep learning; Diabetes; Female; Fibers; Humanities and Social Sciences; Humans; Learning algorithms; Machine Learning; Male; multidisciplinary; Nerve conduction; Nerve conduction velocity; Neural Conduction - physiology; Neural networks; Neural Networks, Computer; Neuropathy; Neurophysiology; Optimization; Partial differential equations; Peripheral neuropathy; Physiology; Propagation; Science; Science (multidisciplinary); Sensory neurons; Signal processing; Signal Processing, Computer-Assisted; Thermodynamic neural networks; Velocity; Wavelet Analysis; Wavelet transform; Wavelet transforms; Thermodynamic neural networks; Wavelet transform; Peripheral neuropathy; Nerve conduction velocity
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-025-08367-7

This paper presents an advanced machine learning (ML) framework for precise nerve conduction velocity (NCV) analysis, integrating multiscale signal processing with physiologically-constrained deep learning. Our approach addresses three fundamental limitations of conventional NCV techniques: (1) oversimplified nerve fiber modeling, (2) temperature sensitivity, and (3) static measurement interpretation. The proposed framework combines: (i) entropy-optimized wavelet analysis for adaptive multiscale signal decomposition, (ii) thermodynamically-regularized neural networks incorporating Arrhenius kinetics, and (iii) stochastic progression models for uncertainty-aware longitudinal tracking. Through data extracted from prior studies in this field, rigorously validated across 1842 patients from 28 medical centers, we demonstrate significant improvements: 23.4% enhancement in motor NCV accuracy ( ) and 28.7% for sensory fibers. The framework maintains physiological interpretability while achieving superior performance through: (a) wavelet-optimized resolution scales (2–8 ms for motor, 0.5–2 ms for sensory fibers), (b) temperature compensation accurate to across 20- , and (c) probabilistic progression tracking with 88.9% treatment response prediction accuracy. This work establishes new standards for ML applications in clinical neurophysiology by rigorously combining biophysical first principles with data-driven learning, offering both theoretical advances and immediate clinical utility for neuropathy diagnosis and monitoring.