AI-Assisted Oscillometry Refines Accuracy of Silicosis Diagnosis

Improving Silicosis Diagnosis through Machine Learning

Silicosis remains a lethal occupational hazard, yet its early detection via oscillometry has long been stalled by the difficulty of interpreting complex lung resistance signals. While traditional spirometry measures volume, it often misses subtle mechanical changes in the distal airways where mineral dust does the most damage. Respiratory oscillometry offers a non-invasive alternative by measuring how lungs respond to pressure waves. However, the raw data is often too complex for manual interpretation. Researchers tested several algorithms, including Explainable Boosting Machines (EBM) and HyperTab, on data from 109 volunteers to automate the Silicosis diagnosis process. The results indicate a significant performance leap:

• Single parameters like resonant frequency achieved an AUC of 0.86.
• HyperTab and EBM models reached a diagnostic accuracy of AUC 0.96.
• Explainable models identified specific feature interactions, providing clinical transparency.

This study suggests that AI-assisted workflows could provide clinicians with a reliable decision-support tool for occupational health screenings. By automating the analysis of complex waveforms, the system reduces the risk of human error in early-stage detection. This study does not solve the challenge of differentiating silicosis from other concurrent interstitial lung diseases in complex clinical presentations.