Echocardiographic risk prediction for heart failure may be enhanced by incorporating advanced signal-derived features, but the added value over conventional parameters remains uncertain. A study published in the International Journal of Cardiology evaluated whether supervised machine learning (ML) applied to strain and tissue Doppler imaging (TDI) curves improves the prediction of incident heart failure (HF) or cardiovascular (CV) death over five years.

The analysis included 2,589 participants from the Copenhagen City Heart Study (CCHS) for model training and 916 participants from the LOOP Study for external validation. A total of 744 signal-derived parameters were extracted from 18 strain curves and 6 TDI curves and evaluated using ML models, which were compared with a baseline ML model trained on 17 conventional echocardiographic parameters.

The primary endpoint was a composite of incident HF or CV death within five years. In the training cohort, 126 individuals (4.9%) met the endpoint, while 59 individuals (6.4%) did so in the validation cohort.

In external testing, models incorporating signal analysis showed improved discrimination compared with the baseline model. The baseline model achieved an area under the curve (AUC) of 0.721 (95% CI 0.639–0.802), while the signal-based model achieved an AUC of 0.790 (95% CI 0.730–0.850; p=0.031), and the combined model achieved an AUC of 0.788 (95% CI 0.727–0.849; p=0.021).

These findings show that ML-based signal analysis of echocardiographic strain and TDI curves provides modest improvement in 5-year prediction of HF or CV death compared with conventional parameters.