Prediction of conduction velocity distribution and motor unit action potential from evoked compound muscle action...

Prediction of motor nerve conduction velocity distribution (CVD) from compound muscle action potential (CMAP), commonly referred to as the inverse problem of nerve conduction, provides a useful computational approach for investigating peripheral neuropathies. Only limited studies have attempted to estimate CVD using continuous parametric models because deterministic optimization approaches, such as gradient-based methods, often face computational challenges when nonlinear CVD functions are considered. In this study, we propose a phenomenological, metaheuristic-based inverse modeling framework for simultaneous estimation of motor unit action potential (MUAP) waveform characteristics and CVD from recorded CMAP signals. The MUAP waveform was modeled using a modified piecewise Hermite polynomial formulation, while the CVD was represented using a modified single Weibull distribution. The mean squared error (MSE) between the recorded and reconstructed CMAP signals was used as the objective function and minimized using genetic algorithm (GA) and grey wolf optimizer (GWO) approaches. The proposed framework efficiently estimated MUAP and CVD parameters with low reconstruction error (MSE < 0.28) for synthetic CMAP signals and demonstrated stable performance when tested on both synthetic andin-vivoCMAP datasets. The maximum computational runtime observed in this study was approximately 2.73 min. Additional analyses showed that GWO provided more consistent convergence than GA and a gradient-based nonlinear least-squares baseline, while noise-robustness testing demonstrated stable reconstruction over multiple signal-to-noise ratio conditions. Group-level comparison of reconstructed CVD parameters further showed distinguishable differences between non-neuropathic and diabetic neuropathy subgroups. The results suggest that the proposed inverse framework provides a computationally efficient and non-invasive approach for analyzing motor nerve conduction characteristics from CMAP recordings and may support quantitative evaluation of peripheral neuropathies.