Clinical Report: Classification of Voice Disorders Through Machine Learning
Overview
This scoping review identifies key barriers to the clinical application of machine learning (ML) in multi-class voice disorder classification. It highlights significant variations in diagnostic labels, data availability, and testing methodologies that hinder robust comparisons and clinical implementation.
Background
Voice disorder classification is essential for advancing voice AI technologies, which have the potential to serve as non-invasive biomarkers for various health conditions. Despite a growing body of research, there remains a gap between experimental findings and clinical application, necessitating a clearer understanding of the barriers to effective ML integration in clinical settings.
Data Highlights
No numerical data available.
Key Findings
80 articles utilized ML techniques for multi-class voice disorder classification out of 10,401 screened.
Significant variation in database selection, diagnostic labels, and input data types was observed.
Inconsistencies in sample size and class imbalance hinder comparisons across studies.
Variations in testing methodologies limit the generalizability of ML models.
The lack of consensus on classification frameworks presents critical barriers to clinical application.
Clinical Implications
Healthcare professionals should be aware of the inconsistencies in ML approaches to voice disorder classification, which may affect diagnostic reliability. Addressing these barriers is crucial for the successful integration of ML technologies into clinical practice.
Conclusion
The review underscores the need for standardized methodologies in voice disorder classification to enhance the clinical applicability of machine learning models. Addressing identified barriers is essential for realizing the potential of voice as a biomarker for systemic diseases.
