Prediction of dysphagia aspiration through machine learning-based analysis of patients’ postprandial voices

Abstract

Abstract Background: Conventional diagnostic methods for dysphagia have limitations such as long wait times, radiation risks, and restricted evaluation. Therefore, voice-based diagnostic and monitoring technologies are required to overcome these limitations. Based on our hypothesis regarding the impact of weakened muscle strength and the presence of aspiration on vocal characteristics, this single-center, prospective study aimed to develop a machine-learning algorithm for predicting dysphagia status (normal, and aspiration) by analyzing postprandial voice limiting intake to 3cc. Methods: This study was a single-center, prospective cohort study, conducted from September 2021 to February 2023, at the Seoul National University Bundang Hospital. A total of 204 participants were included, aged 40 or older, comprising 133 without suspected dysphagia and 71 with dysphagia-aspiration.Voice data from participants were collected and used to develop dysphagia prediction models using the Audio Spectrogram Transformer process with MobileNet V3. Male-only, female-only, and combined models were constructed using 10-fold cross-validation. Through the inference process, we established a model capable of probabilistically categorizing a new patient's voice as either normal or indicating the possibility of aspiration. Results: The pre-trained models (mn40_as and mn30_as) exhibited superior performance compared to the non-pre-trained models (mn4.0 and mn3.0). The best-performing model, mn30_as, which is a pre-trained model, demonstrated an average AUC across 10 folds as follows: combined model 0.7879 (95% CI 0.7355-0.8403; max 0.9531), male model 0.7787 (95% CI 0.6768-0.8806; max 1.000), and female model 0.7586 (95% CI 0.6769-0.8402; max 0.9132). Additionally, the other models (pre-trained; mn40_as, non-pre-trained; mn4.0 and mn3.0) also achieved performance above 0.7 in most cases, and the highest fold-level performance for most models was approximately around 0.9. Conclusions: This study suggests the potential of using simple voice analysis as a supplementary tool for screening, diagnosing, and monitoring dysphagia aspiration. By directly analyzing the voice itself, this method enables simpler and more remarkable analysis in contrast to conventional clinical evaluations. The postprandial voice-based prediction model holds implications for improving patient quality of life and advancing the development of non-invasive, safer, and more effective intervention methods. Trial registration: This study was approved by the IRB (No. B-2109-707-303) and registered on clinicaltrials.gov (ID: NCT05149976).