Medical-AI trial: detecting congenital heart disease from heart sounds recorded on a smartphone mic — assessing an algorithm (ausculto) (NCT07376785)

Abnormalities of the heart valves or blood flow can appear as a murmur in the heart sounds heard through a stethoscope. Many childhood murmurs are innocent and need no treatment, but some signal congenital heart disease (a heart defect present from birth), and telling them apart takes expertise.

If AI could distinguish abnormalities from heart sounds recorded on a smartphone microphone, it might help screen the heart even where stethoscopes and specialists are scarce.

This study assesses the performance of that smartphone heart-sound software, ausculto, as an observational study. Per the registry summary, ausculto is a set of computer algorithms intended to analyze heart sounds recorded from a smartphone built-in microphone and find abnormal sounds. The primary outcome is telling murmurs associated with congenital heart disease from normal sounds and innocent murmurs.

Participants have their heart sounds recorded during normal hospital attendance, and researchers manually annotate the recordings to build a database for future training and testing of medical AI. The size is 220, conducted by invitation.

What stands out is that the study does not claim the effectiveness of a finished AI; alongside assessing performance, its focus is building a high-quality annotated database. The foundation of medical AI is good training data, and having experts label heart sounds gathered on a familiar smartphone supports the next stage of AI development.

It could widen congenital-heart-disease screening with accessible devices, but this study is at the performance-and-data stage; it does not replace diagnosis or establish effectiveness.