The CONVALV mobile AI pipeline demonstrated technical feasibility for the automated analysis of non-invasive cardiac auscultation using public reference datasets.
Does the CONVALV AI pipeline accurately classify normal and abnormal heart sounds from acoustic recordings?
The CONVALV mobile-to-AI pipeline demonstrates high accuracy and sensitivity for automated binary classification of heart sounds, showing technical feasibility for integration into digital telemedicine platforms.
Cardiac auscultation is a widely used non-invasive tool in cardiovascular assessment, although its diagnostic reliability depends on the practitioner's experience and the conditions under which the auscultation is performed. Recent advances in artificial intelligence and mobile technologies allow for mitigating these limitations through automated, scalable, and accessible screening systems. This paper presents CONVALV, an application integrated into a functional web platform for the automatic analysis of heart sounds acquired via consumer devices. The system implements a mobile-to-AI pipeline. This architecture encompasses acoustic signal conditioning, hybrid extraction of physiologically interpretable features, and inference using deep learning models, providing rapid and reproducible screening results accessible through a digital interface. The proposed architecture is optimized for execution in computationally limited environments and geared towards future integration into telemedicine scenarios. Evaluation is performed using public reference datasets under rigorous subject-level validation protocols. Although CONVALV is presented as a research prototype and not a medical device, the results obtained demonstrate the technical feasibility of integrating automated analysis of non-invasive cardiac auscultation into digital platforms that support clinical decision-making.
Robert Martin (Thu,) conducted a other in Cardiovascular assessment. CONVALV mobile AI pipeline was evaluated on Technical feasibility. The CONVALV mobile AI pipeline demonstrated technical feasibility for the automated analysis of non-invasive cardiac auscultation using public reference datasets.