Voice as a Biomarker to Detect Acute Decompensated Heart Failure: Pilot Study for the Analysis of Voice Using Deep Learning Models

BackgroundAcute decompensated heart failure (ADHF) is a systemic congestion state requiring timely management. Admission for ADHF is closely related to the readmission and post-discharge mortality in patients, which makes it imperative to detect ADHF in its early stage. MethodsPatients with ADHF needed admission were eligible for enrollment, and those with respiratory infection, sepsis, lung/vocal cord disease, acute coronary syndrome, or serum creatinine>3mg/dL were excluded. A total of 112 patients were enrolled between July, 2020 and December, 2022. Voice was recorded two times: at admission for ADHF, and at discharge. Patients were asked to phonate five Korean vowels ( a/e/i/o/u) for 3 seconds each, and then to repeat the sentence daehan minkook manse five times. Low-level audio features were extracted for classification. Then, Mel-Spectrogram was extracted from waveform and used as input features of the deep learning-based classification models. Two kinds of the deep learning-based classification models, convolutional neural networks and Transformer, were adapted for the further analysis. ResultsFor 100 patients in the final analysis, we randomized patients into two mutually exclusive groups: a training group (n=88) and a test group (n=12). In the analysis with low-level audio features, harmonics-to-noise ratio and Shimmer showed classification potential. Then, deep learning models were trained to classify whether certain voice belongs to ADHF state or recovered state. We treated it as a binary classification task, and the best performing model achieved a classification accuracy of 85.11% with DenseNet201. The classification accuracy was improved as 92.76% with ViT-16-large after inputting additional classic features of heart failure. With adding the low-level audio features in a training process, classification task accuracy was improved in DenseNet201 for about 2%. ConclusionsOur results proposed the clinical possibility of voice as a useful and noninvasive biomarker to detect ADHF in its early stage.