The Rise in Artificial Intelligence and Machine Learning Models to Screen for Cleft-Related Velopharyngeal...

Area of Science:

• Medical Informatics
• Speech Pathology
• Artificial Intelligence

Background:

• Velopharyngeal dysfunction (VPD) is a common complication in patients with cleft palate, affecting speech intelligibility.
• Accurate detection of VPD is crucial for timely intervention and improved patient outcomes.
• Traditional methods for VPD assessment can be subjective and time-consuming.

Purpose of the Study:

• To systematically review the existing literature on the application of artificial intelligence (AI) and machine learning (ML) models for detecting velopharyngeal dysfunction (VPD).
• To evaluate the performance and generalizability of AI/ML models in identifying speech abnormalities associated with VPD in cleft palate patients.

Main Methods:

• A systematic review was conducted following PRISMA guidelines, searching databases like PubMed, EMBASE, ProQuest, and Google Scholar.
• Machine learning models were trained and validated on speech features (e.g., MFCCs, CQCCs) from a large dataset of participants with and without VPD.
• Performance metrics including accuracy, precision, recall, F1-score, sensitivity, specificity, and Pearson correlation coefficient were analyzed.

Main Results:

• 34 articles met the inclusion criteria, with Support Vector Machines being the most common model (47.1%).
• AI/ML models demonstrated promising average performance: accuracy 82.9%, precision 86.7%, F1-score 0.88, sensitivity 80.5%, specificity 82.2%.
• Only 8.8% of studies performed external validation, highlighting a significant limitation.

Conclusions:

• AI/ML models show potential for detecting VPD in cleft palate patients, offering encouraging performance metrics.
• Inconsistent reporting standards, reliance on engineered speech features, and a lack of external validation hinder the generalizability of current models.
• No AI/ML model has yet achieved clinical deployability for VPD detection.