An AI-assisted multimodal diagnostic test for heart failure

Heart failure (HF) – inability of the heart to provide a sufficient amount of blood for normal body function, represents an immense societal burden due to a high mortality rate (50% of patients die within 5 years from diagnosis), global pandemic proportions and high healthcare expenditures. When diagnosed in an early phase, HF can be arrested or even reversed, while a late diagnosis inevitably leads to the fatal outcome. However, 70% of cases are diagnosed only upon a series of tests: ECG, biochemical and echocardiographic, in secondary care [1], leaving many patients not treated timely or at all. Hence, coordinated efforts are underway across medical and engineering communities to establish new tests suitable for primary care and screening. We joined these efforts through the project SensSmart, focused on the development of multimodal sensor technology for timely HF diagnosis. SensSmart technology relies on non-invasive synchronized measurement of electrical and mechanical processes in the cardiovascular system, including the electrocardiogram, heart sounds and movement, and arterial pulsations. Rooted in earlier investigations [2], our technology leverages modern electrical, mechanical and photonic sensing [3], along with AIbased data processing [4], to provide an HF diagnostic test suitable for primary care. We stress the enabling role of deep learning, noting that its capability to simultaneously process numerous features of different physiological signals and their interdependences surpasses that of humans. The algorithm was trained on data from the SensSmart clinical study conducted from Oct 2023 to Sep 2025 at two centres of the University Clinical Centre of Serbia. Based on the preliminary analysis, the system performs binary HF classification with a sensitivity of 90% and an F1-score of 87%, while demonstrating a clear trend of performance improvement with the addition of new modalities. During the talk, I will provide details of the sensing system, classification algorithm and clinical training set and outline future system upgrades.