Work

Cardiac segmentation for ventricular tachycardia radioablation and radiotherapy patients with cardiac implants

Year: 2026

Type: preprint

Abstract: Purpose: Deep-learning (DL) auto-segmentation enables cardiac substructure (CS) delineation for advanced cardiac-sparing radiation therapy (RT). Patients presenting for RT may have implantable devices such as cardioverter-defibrillators (ICDs), which induce metal artifacts. Furthermore, stereotactic body RT (SBRT) is used to treat ventricular tachycardia (VT), a population frequently presenting with ICDs. We apply a DL model for RT patients with ICDs or undergoing VT-SBRT, enabling auto-segmentation and substructure-level dose analysis. Methods: Nineteen patients with ICDs who underwent CT simulation (CT-SIM) and received VT-SBRT were evaluated. A previously validated multi-modal nnU-Net-based segmentation method was retrained using a highly-curated CT-SIM cohort, supplemented with ICD images for training (ICD-specific, 8; total, 33) and validation (ICD-specific, 1; total, 6), with a hold-out ICD test set (n=10), to predict 20 CS including whole heart (WH), chambers, great vessels, coronary arteries, valves, and conduction nodes. Additional CT-SIM datasets included cancer-RT patients with (n=9) and without (n=10) ICDs evaluate generalizability. The retrained ICD-supplemented model was compared against a non-ICD model using Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95), and Wilcoxon signed-rank test (p<0.05). Dosimetric evaluation (Dmean, maximum (D0.03cc)) was performed for VT-SBRT CS. Results: For VT-SBRT, the ICD-supplemented model achieved average DSC/HD95 of 0.71±0.19/10.7±22.5mm respectively, outperforming (p<0.05) the non-ICD model. On CT-SIM images with ICDs, the model achieved an average DSC/HD95 of 0.70±0.22/7.9±5.1mm respectively, outperforming (p<0.05) the non-ICD model. On CT-SIM images without ICDs, the model achieved an average DSC of 0.74±0.17 (Non-ICD model, DSC, 0.75±0.16, p<0.05) and average HD95 of 5.5±3.1mm (Non-ICD model, HD95, 5.3±2.8mm). For VT-SBRT, CS dose depended on target location, with higher relative dose delivered to the WH, left/right ventricles, tricuspid valve, atrioventricular node, and left anterior descending artery than other substructures. Conclusion: ICD-supplemented training improved CS segmentation despite metal artifacts without meaningfully degrading non-ICD performance, supporting auto-segmentation and substructure-level dosimetry in ICD populations.

Source: PubMed Central

DOI: 10.48550/arxiv.2609.36367