Left ventricular ejection fraction (LVEF) is the cornerstone parameter for cardiotoxicity surveillance in patients receiving cardiotoxic cancer therapies — and the primary criterion by which cardiac dysfunction is defined in cardio-oncology guidelines. Yet manual LVEF assessment is operator-dependent and subject to significant inter-observer variability, limiting the reliability of serial monitoring in longitudinal follow-up programmes. This task challenges teams to develop automated, reproducible LVEF estimation directly from echocardiography video, enabling consistent cardiac surveillance at every timepoint across a patient's treatment journey.

LV dysfunction — characterised by impaired global longitudinal strain (GLS ≥ −16%, per ASE/EACVI consensus) — represents an early manifestation of subclinical cardiotoxicity, detectable weeks to months before overt LVEF decline. Identifying it promptly is critical: it marks the window in which cardioprotective intervention is still effective and therapy modification remains a viable option. This task challenges teams to classify LV dysfunction from longitudinal echocardiography, turning automated cardiac assessment into a concrete clinical decision at each follow-up timepoint.

Predicting cardiotoxicity from baseline imaging — before therapy begins — represents the earliest and most consequential point of clinical intervention: the only moment at which cardiac damage can be prevented entirely, rather than managed after the fact. High-risk patients identified at baseline can receive cardioprotective co-treatment or modified regimens from the outset, while low-risk patients avoid unnecessary intervention. This task challenges teams to predict future cardiotoxicity from pre-treatment echocardiography alone, enabling preventive cardio-oncology care at the point where it matters most.