Electrophysiological-based automatic subgroups diagnosis of patients with chronic dysimmune polyneuropathies.

Background: Chronic Dysimmune Polyneuropathies (CDP) encompass a group of conditions characterized by autoimmune etiology targeting myelin and/or axonal components. Subgroups include Chronic Inflammatory Demyelinating Polyneuropathy (CIDP) and paraproteinemic neuropathies associated with IgM monoclonal gammopathy of undetermined significance, with anti-MAG antibodies (IgM-MGUS anti-MAG+) and without anti-MAG antibodies (IgM-MGUS anti-MAG-). Their identification is crucial for determining the most suitable treatment options, yet it poses significant challenges. In this study, an electrophysiological-based automatic classification through machine learning models is proposed.

Methods: This study included 67 patients, 29 diagnosed with CIDP, 20 with polyneuropathy associated with IgM-MGUS anti-MAG+, and 18 with CIDP-like polyneuropathy associated with IgM-MGUS anti-MAG-. Five different classification algorithms based on electrophysiological data (conduction velocity, latency, and amplitude of sensory and motor responses from different nerves) were implemented to classify three types of neuropathies and identify discriminative neurographic parameters.

Results: The best performance in stratifying the three classes was achieved by Random Forest in terms of both validation and test accuracy (86.5% and 80.6%). The predictor analysis on the best-performing model revealed the significance of F-wave latencies, distal latencies, and conduction velocities for group discrimination.

Conclusions: The study is the first to apply computational methods to identify electrophysiological parameters most frequently altered in different forms of polyneuropathy, to support clinical diagnosis and decision-making.

Supplementary Information: The online version contains supplementary material available at 10.1186/s12984-025-01685-x.