Optimizing Nociceptive Flexion Reflex (NFR) Scoring Criteria by Adjusting for Noise and Reflex Properties and...

To reanalyze scoring criteria for automatic detection of nociceptive flexion reflexes (NFRs) in electromyography (EMG) recordings and to improve detection accuracy by accounting for multiple characteristics of the recordings, such as baseline noise level or sampling rate.

Single scoring criteria for the NFR were reanalyzed and validated against an independent data set. To account for influences on the single scoring criteria, such as the baseline noise, multivariate classification models were derived.

Reanalysis of single scoring criteria yielded significantly lower threshold values than previously reported. The threshold value of the best-performing single scoring criterion, the NFR Interval Peak z score, was found to be strongly dependent on the level of baseline noise and the EMG sampling rate. Multivariate classification models could reduce the number of incorrectly classified recordings in an independent data set by 25% to 37% compared with the best-performing single scoring criterion.

The automatic detection of reflex responses in electromyograms can be significantly improved by including multiple reflex, baseline, and EMG characteristics into a classification model. These findings should help to improve the accuracy of currently used standard measurement algorithms and algorithms engineered toward specific properties, such as short measurements or less induced pain for the patients.