The Rise in Artificial Intelligence and Machine Learning Models to Screen for Cleft-Related Velopharyngeal Dysfunction: A Systematic Review

Comprehensive Summary

This article systematically reviewed the growing body of literature evaluating artificial intelligence and machine learning approaches for detecting velopharyngeal dysfunction (VPD) in patients with cleft palate. The authors analyzed 34 studies comprising 3,967 participants and more than 92,000 speech training samples. Most studies used acoustic speech features to train models capable of identifying hypernasality and other speech abnormalities associated with VPD. The most frequently used algorithms were support vector machines,neural networks, and deep neural networks. A common theme across many studies was that AI/ML systems show generally favorable diagnostic performance, with average accuracy, sensitivity, and specificity values over 80%. However, the review noted major limitations relating to methodology that undermine clinical generalizability, including inconsistent reporting standards, heterogeneity in recording conditions, and relying too much on internally validated datasets. Importantly, none of the studies used cleft palate patients without VPD as controls which raised concerns that models may learn to recognize cleft-related speech patterns rather than true pathological dysfunction. The authors concluded that although AI-based screening for VPD shows substantial promise, no currently available model has sufficient reproducibility or external validation to support widespread clinical use.

Outcomes and Implications

When used in the clinic, VPD screening systems could dramatically expand access to specialized speech pathology services, particularly in low- and middle-income countries where trained speech-language pathologists are scarce. Earlier and more accessible detection of VPD could improve referral timing and standardize assessment across institutions. The findings of the article also suggest that language-agnostic acoustic biomarkers may eventually allow for the development of universally adaptable screening tools. Overall, the article positions AI not as a replacement for speech-language pathologists, but as a scalable adjunct capable of extending specialized cleft care to underserved populations if future studies address the methodological weaknesses identified in the current literature.