Mobile phone auscultation combined with nonlinear dynamics analysis detected aortic stenosis with a test set sensitivity of 92.3% and AUC of 0.872 compared to echocardiogram diagnosis.
Observational (n=248)
Open label
No
Does mobile phone auscultation analyzed via nonlinear dynamics accurately detect aortic stenosis in adults ≥50 years old?
Auscultation recordings collected via unmodified mobile phones and analyzed with nonlinear dynamics software demonstrated feasibility and high accuracy for detecting isolated aortic stenosis.
Effect estimate: AUC 0.872
Background The prevalence of valvular heart disease is increasing. Early detection remains poor as screening relies on front line detection of audible or symptomatic disease and confirmation requires specialized echocardiography. Methods We conducted a single center, observational pilot study. Eligible subjects were stratified into groups based on echocardiographic findings. In addition to chart extraction of demographics, medical history, and echocardiographic parameters, each subject underwent three auscultation recordings that were analyzed via computational nonlinear dynamics to extract features and construct predictors without fitting or weighting. Predictors were used to create logistic regression binary classification models. Training and test set performance was reported for each model with a focus on area-under-the-curve and sensitivity as the primary benchmarks. Results We analyzed the recordings of 248 subjects, median age 73 years, 43.6% female, 99% White. All recordings were chaotic and of low dimensionality. Personnel and subject collected recordings had a normalized mutual information entropy of 1.0, indicating they shared the same information and could be interchangeable for model development. Three models for aortic stenosis met predetermined metrics, with the best performing model reporting an AUC of 0.872 and a sensitivity of 0.923. Mitral regurgitation models were explored but limited by sample size. Conclusions This study established the feasibility of two innovative approaches, by combining the sound recordings collected from unmodified mobile phones with analysis via nonlinear dynamics software. This work has the potential to improve valvular heart disease detection by overcoming barriers that remain for current standards of care and emerging artificial intelligence solutions.
Close et al. (Wed,) conducted a observational in Adults ≥50 years old with isolated, unrepaired aortic stenosis of any severity (mild, moderate, or severe) or isolated, unrepaired mitral regurgitation (moderate or severe) or other structural heart disease or healthy controls (n=248). Mobile phone auscultation with unmodified microphones analyzed by nonlinear dynamics software vs. No mobile phone auscultation (echocardiogram gold standard) was evaluated on Binary diagnosis of aortic stenosis as determined by cardiologist based on echocardiogram (AUC 0.872). Mobile phone auscultation combined with nonlinear dynamics analysis detected aortic stenosis with a test set sensitivity of 92.3% and AUC of 0.872 compared to echocardiogram diagnosis.