Prediction of dysphagia aspiration through machine learning-based analysis of patients' postprandial voices.
Jung-Min Kim, Min-Seop Kim, Sun-Young Choi and 1 others
PMID 38555417WHAT IT FOUND
A machine learning model analyzing post-meal voice recordings distinguished aspiration from normal swallowing with an average accuracy score of 0.83.
This offers a potential non-invasive screening tool, though it requires further validation before clinical use.
Key findings
01The combined gender model achieved an average Area Under the Curve (AUC) of 0.8361, indicating good ability to distinguish between normal and aspiration cases.
02The model's sensitivity was 77.80% and specificity was 77.52%, meaning it correctly identified most aspiration cases and most normal cases.
03Pre-trained models performed better than non-pre-trained models, likely due to the limited size of the clinical voice dataset.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The study did not use an independent validation set, relying instead on cross-validation, which may overestimate performance. There was a significant gender imbalance in the aspiration group (52 men vs 18 women), leading to lower sensitivity in the female model. Healthy volunteers and patients were recorded in different environments (soundproof room vs VFSS room), introducing potential noise bias. Diet types were not standardized across all participants, which could affect voice quality. The model is a 'black box'; researchers could not determine which specific voice features drove the predictions.
Declared interests
The study was funded by the National Research Foundation of Korea and the SNUBH Research Fund. The authors declared no competing interests.
The easy way to misread this
Do not assume this tool is ready for clinical use. The study lacked an independent validation set, and the model's performance varied significantly by gender, with lower sensitivity in women. It is a proof-of-concept for a future device, not a current diagnostic standard.