VentrEX : An Anatomically Guided Deep Learning Pipeline for Ventricular Segmentation in Cine Cardiac MRI

Background : Automated segmentation of the left and right ventricles (LV) and (RV) in cine cardiac MRI (CMR) underpins reliable volumetry and mass estimation. However, papillary muscles and trabeculae (PM/T) introduce clinically meaningful variability and exacerbate cross-dataset domain shift. Methods : We present VentrEX, an anatomically guided pipeline. The core segmenter, VentrEX-Seg is a 3D encoder-decoder with {parallel channel-spatial attention} and a {Transformer} bottleneck. Training is performed exclusively on ACDC. A lightweight PM/T module automatically extracts papillary and trabecular burden and standardizes cavity volumes. External evaluation is {zero-shot} (no fine-tuning) on Sunnybrook (LV) and MM-WHS MRI (RV). We report Dice, HD95 (mm); for volumetry we use Bland-Altman analyses (LV and RV volumes). Attention/Grad-CAM visualizations support interpretability. Results : On ACDC, VentrEX achieved higher Dice and lower boundary error than U-Net, nnU-Net, CBAM, and VentrEX-Seg. Zero-shot performance was preserved externally (e.g., Sunnybrook LV Dice~0.9053, HD95~4.95\,mm; MM-WHS RV Dice~0.9236, HD95~6.61\,mm). PM/T standardization reduced volumetric bias and narrowed limits of agreement in Bland-Altman analyses. Qualitative overlays and 3D reconstructions showed fewer PM/T "leaks" and anatomically plausible borders across ED/ES. Conclusions : Single-source training with dual zero-shot external tests demonstrates robustness under domain shift, while explicit PM/T modeling reduces volume bias and improves reproducible volumetry. The combination of parallel attention and a Transformer bottleneck enables accurate, transparent cine-CMR segmentation across datasets.