AI-driven speech biomarkers for disease diagnosis and monitoring: a systematic review and meta-analysis

Comprehensive Summary

This study, conducted by Yang et al., examines literature on the use of speech biomarkers and artificial intelligence in disease diagnosis and monitoring. They used data from six databases: PubMed, Embase, Scopus, Web of Science, PsycINFO, and IEEE Xplore, covering studies published from database inception to May 2024. They collected two forms of data: qualitative data, focusing on changes in speech patterns or disease types that the speech biomarkers picked up, and quantitative data, focusing on the sensitivity and specificity of the speech biomarkers. Meaning they analyzed the performance of speech biomarkers in identifying the correct disease by monitoring changes in patients' voices. They found that their pooled sensitivity and specificity for diagnostic models were between 0.77 and 0.85. Furthermore, they studied the variations in the recordings that were significant contributors to the interpretation of the disease, such as recording device, language, speech task, speech feature,s and algorithm selection. Overall, they found that speech biomarkers show promise for diagnosing and monitoring diseases, which could be particularly useful in home-based observations via smartphones. However, there is a high risk of bias in many studies, especially in patient selection and index test interpretation, which limits the strength of current evidence.

Outcomes and Implications

This research is important because it aims to utilize AI to help clinicians diagnose and treat many neurological and psychiatric conditions. AI focuses on the patient’s speech, something the clinician can miss, to find a pattern that alerts the provider that there is a possible disease. While this technology could be really helpful to clinicians in detecting the disease earlier, they still need to be wary of biases and perform their own examinations before coming to conclusions.