From Support Vector Machines to Neural Networks: Advancing Automated Velopharyngeal Dysfunction Detection in Patients...

Area of Science:

• Speech Pathology
• Artificial Intelligence
• Machine Learning

Background:

• Intelligible speech is crucial after cleft palate repair.
• Velopharyngeal dysfunction (VPD) significantly impacts speech outcomes.
• The prevalence and impact of VPD in low- and middle-income countries (LMICs) are not well understood.

Purpose of the Study:

• To explore the use of AI and machine learning (ML) for automatic detection of VPD from speech samples.
• To develop and test a deep learning model for VPD identification in patients with cleft palate.
• To assess the potential of AI-driven tools to enhance VPD care in resource-limited settings.

Main Methods:

• Development of a neural network-based, self-supervised deep learning ML model.
• Utilized a cohort of 60 patients (30 with VPD, 30 controls) with cleft palate.
• Trained and tested the model on approximately 8000 audio speech samples.

Main Results:

• The ML model achieved perfect accuracy, precision, recall, and F1 scores (1.0) in detecting VPD.
• The model demonstrated high performance on both augmented and unaugmented datasets.
• Initial results indicate a strong capability of ML in identifying speech patterns associated with VPD.

Conclusions:

• Promising results support the potential of ML models for detecting VPD in speech.
• Further research is needed to address confounding factors and enable multilingual analysis.
• Clinical implementation of such AI models could significantly improve VPD care for cleft palate patients in LMICs.