Machine Learning for Automatic Detection of Velopharyngeal Dysfunction: Proof of Concept

• DOI: 10.1097/01.gox.0001024432.49195.f3
• Publication Date: 2024-06-03T10:02:13Z
• Authors: Nicholas O’Sick; Claiborne Lucas; Ricardo Torres-Gu...; Andrew J. James; Scott Corlew; Amy Stone; Maria E. Powell; Michael Golinko; Matthew E. Pontell
• Keywords: Proof of concept; Burden of Proof

Background: Even after palatoplasty, the incidence of velopharyngeal dysfunction (VPD) can reach 30%; however, these estimates arise from high-income countries (HICs) where speech-language pathologists are part standardized cleft teams. The VPD burden in low- and middle-income (LMICs) is unknown. This study aims to develop a machine learning model that detect presence using audio samples alone. Methods: Case control were obtained by institutional publicly available sources. A was built Python software. Results: 110 initial used test train re-tested format conversion file deidentification. Each sample tested 5 times yielding precision 100%. Sensitivity 92.73% (95% CI 82.41%-97.98%) specificity 98.18% 90.28%-99.95%). 113 prospective samples, which had not yet interacted with model, then tested. Precision again 100% sensitivity 88.89% 78.44%-95.41%) 66% 51.23%-78.79%). Discussion: affects nearly patients unrepaired overt soft palatal clefts 30% who have undergone palatoplasty. render unintelligible, thereby accruing significant psychosocial morbidity. LMICs unknown, likely exceeds HICs. ability access phone-based screening could expand diagnostic, potentially therapeutic, modalities for an innumerable amount world-wide suffer VPD.