To identify the key barriers to the clinical application of Machine Learning (ML) in multi-class voice disorder classification, emphasizing the importance of overcoming these barriers for effective implementation.
Key Findings:
80 articles used ML techniques for multi-class classification out of 10,401 screened, highlighting the need for more focused research.
Significant variation in database selection, diagnostic labels, input data types, and classification techniques was observed, which complicates the path to clinical application.
Inconsistencies hinder robust comparisons and identification of state-of-the-art solutions, limiting the potential for effective clinical use.
Interpretation:
Variations in classification tasks and methodologies limit comparability and undermine the generalization of ML models, posing challenges for their integration into clinical practice.
Limitations:
Lack of consensus on diagnostic labels and classification frameworks, which complicates communication among practitioners.
Variability in sample size and class imbalance across studies, which may skew results and affect reliability.
Differences in testing methods and validation approaches, which can lead to inconsistent findings and hinder reproducibility.
Conclusion:
The review highlights critical barriers to the clinical application of ML in voice disorder classification, emphasizing the urgent need for standardization in diagnostic labels and methodologies to facilitate effective implementation.
