A study published in Muscle & Nerve on June 2, 2026, evaluated a hybrid artificial intelligence system designed to automate electrodiagnostic (EDX) reporting. The system combines a rule-based extractor with a constrained large language model (LLM) to convert structured nerve conduction and electromyography measurements into standardized clinical reports. Researchers conducted a retrospective analysis using 332 reports from two different institutions, including 241 primary-care and 91 tertiary-care examinations. The system first generated a summary of abnormal findings using laboratory-specific cutoffs, followed by the LLM drafting diagnostic interpretations by selecting from a predefined label inventory.

To measure performance, the team utilized a Coverage–Localization Index (CLI), which assesses both the correct selection of diagnostic labels and the accuracy of site or level localization. The results showed a mean total agreement of 96.7% (95% CI 95.1–98.1) between the system's output and human clinician interpretations. While agreement was high overall, the system showed lower performance on more complex examinations. The rule-based component successfully captured all defined abnormal findings, and the LLM produced consistent, standardized drafts. The study authors noted that because the evaluation was limited to abnormal, in-scope reports, future research must validate the system on consecutive routine clinical streams, including normal cases, to fully assess workflow integration and clinician acceptability.

A study published in Muscle & Nerve on June 2, 2026, evaluated a hybrid artificial intelligence system designed to automate electrodiagnostic (EDX) reporting. The system combines a rule-based extractor with a constrained large language model (LLM) to convert structured nerve conduction and electromyography measurements into standardized clinical reports. Researchers conducted a retrospective analysis using 332 reports from two different institutions, including 241 primary-care and 91 tertiary-care examinations. The system first generated a summary of abnormal findings using laboratory-specific cutoffs, followed by the LLM drafting diagnostic interpretations by selecting from a predefined label inventory.

To measure performance, the team utilized a Coverage–Localization Index (CLI), which assesses both the correct selection of diagnostic labels and the accuracy of site or level localization. The results showed a mean total agreement of 96.7% (95% CI 95.1–98.1) between the system's output and human clinician interpretations. While agreement was high overall, the system showed lower performance on more complex examinations. The rule-based component successfully captured all defined abnormal findings, and the LLM produced consistent, standardized drafts. The study authors noted that because the evaluation was limited to abnormal, in-scope reports, future research must validate the system on consecutive routine clinical streams, including normal cases, to fully assess workflow integration and clinician acceptability.