Matton, Katie & Balaji, Purvaja & Ghasemzadeh, Hamzeh & Cooper, Jameson & Mehta, Daryush & Van Stan, Jarrad & Hillman, Robert & Picard, Rosalind & Guttag, John & Abulnaga, Mazdak. (2025). Classifying Phonotrauma Severity from Vocal Fold Images with Soft Ordinal Regression. 10.48550/arXiv.2511.09702.

Abstract

Phonotrauma refers to vocal fold tissue damageresulting from exposure to forces during voic-ing. It occurs on a continuum from mild tosevere, and treatment options can vary basedon severity. Assessment of severity involves aclinician’s expert judgment, which is costly andcan vary widely in reliability. In this work, wepresent the first method for automatically clas-sifying phonotrauma severity from vocal foldimages. To account for the ordinal nature ofthe labels, we adopt a widely used ordinal re-gression framework. To account for label uncer-tainty, we propose a novel modification to ordi-nal regression loss functions that enables themto operate on soft labels reflecting annotatorrating distributions. Our proposed soft ordi-nal regression method achieves predictive per-formance approaching that of clinical experts,while producing well-calibrated uncertainty es-timates. By providing an automated tool forphonotrauma severity assessment, our work canenable large-scale studies of phonotrauma, ulti-mately leading to improved clinical understand-ing and patient care.