Changing the Rhythm: How the Hearts Trial Is Redefining Cardiac Safety In Cancer Treatment
Abstract
Purpose
Cardiotoxicity is a devastating complication of cancer treatment. Notably, the heart is complex and dose to cardiac substructures (CS) have been strongly linked to radiation-induced cardiac morbidity. The physicist-led, NIH-funded “THE cARdiac Radiation Therapy Sparing (HEARTS)” trial hypothesizes that applying advanced deep learning (DL) techniques and MRI-guided adaptive radiation therapy (MRgART) in thoracic cancer patients will yield meaningful changes in cardiotoxicity endpoints.
Methods
HEARTS is a prospective, randomized, phase II interventional trial of 60 thoracic cancer patients comparing MRgART with CS sparing (MR-linac planning/treatment, gated on end-inhalation) to Standard of Care (linac-based, cone-beam CT using whole heart endpoints, free-breathing delivery with internal target volume). To determine dosimetric eligibility (whole heart V25Gy>10%), clinical datasets (e.g, PET-CT, diagnostic CT) undergo rapid tumor/OAR nnU-Net-based segmentation and 5-fold cGAN DL dose prediction. All subjects undergo baseline cardiac MRI including a novel 5DMRI sequence to quantify cardiac and respiratory CS motion for margin assessment. Experimental arm (MRgART) patients have 20 CS segmented via nnU-Net, used for cardiac-spared treatment planning. The primary endpoint is longitudinal changes (3, 6 months post-treatment) from baseline in left ventricular ejection fraction and secondary endpoints include functional cardiac MRI, quality of life, heart rhythm, clinical toxicities, and exploratory blood biomarker measurements.
Results
IDE exemptions were granted for developed DL and 5DMRI technologies with study activation in August, 2025. Upon confirmation of gross eligibility, HEARTS candidates underwent DL dose prediction and automated structured eligibility reporting within 2 hours. The 5DMRI processing pipeline was successfully implemented to characterize CS excursion for cardio-respiratory (free-breathing) and cardiac (breath-hold) treatment conditions. Once data is transferred, CS prediction is performed in ~10 seconds with automated dosimetric endpoint evaluation performed following fractionation schemes.
Conclusion
Early technical success of our novel, physics‑driven clinical trial underscores the feasibility and transformative potential of our cardiac‑sparing strategies in thoracic radiotherapy.