New research: Near-boundary double-labeling-based classification

In a clinical study aiming to validate a machine learning–based algorithm for mandibular jaw movement signals on 289 patients, Dr. Jean Benoit Martinot et al. applied the innovative Near Boundary Labelling (NBL) approach.

Key findings

This innovative approach allows to assign individuals with apnea-hypopnea index (AHI) in predefined near-boundary zones to two different categories of AHI grades of severity. The researchers postulated that the risk of AHI-based severity misclassification due to inter-human PSG rating could be reduced.

The four-way Cohen’s Kappa coefficient was improved from 0.80 to 0.86. The NBL rule had also a positive impact on the accuracy in OSA severity grading. The F1 score, indicating the harmonic mean of precision and sensitivity was improved from 0.93 to 0.96 for detecting moderate OSA, and from 0.87 to 0.88 for severe OSA. 

Read the study: https://lnkd.in/eDA-mGn8](https://lnkd.in/eDA-mGn8