Published Paper Details:

Automated segmentation of cardiac structures from cine magnetic resonance imaging (cMRI) is a cornerstone of cardiovascular diagnostics, yet the field remains without a unified evaluation framework that rigorously examines the interplay between backbone architecture, adversarial supervision, and optimizer design. More critically, no systematic investigation has explored whether quantum-inspired computational paradigms -- specifically Quantum-Behaved Particle Swarm Optimization (QPSO) and quantum-classical hybrid Quantum Convolutional Neural Networks (QCNN) -- can deliver meaningful performance advantages in this domain. This paper closes both gaps through an exhaustive benchmarking study of thirty deep learning configurations on the publicly available Automated Cardiac Diagnosis Challenge (ACDC) dataset, the gold standard for cardiac Magnetic Resonance Imaging (MRI) segmentation evaluation. Seven backbone architectures are interrogated -- standard encoder-decoder network with skip connections (UNet), Residual Network with 34 layers encoder-based UNet (ResNet34-UNet), lightweight Custom UNet (CustomUNet), Shifted Window Transformer-based UNet (SwinUNet), cardiac-specialized segmentation network (CardioSeg), Quantum Convolutional Neural Network (QCNN), and Attention-gated UNet (AttentionUNet) -- each systematically paired with training strategies spanning gradient-based optimizers including Adaptive Moment Estimation (Adam), Stochastic Gradient Descent (SGD), Root Mean Square Propagation (RMSProp), Nesterov-accelerated Adaptive Moment Estimation (NAdam), and Adaptive Gradient Algorithm (Adagrad), alongside classical Particle Swarm Optimization (PSO), Quantum-Behaved Particle Swarm Optimization (QPSO), and Generative Adversarial Network (GAN)-augmented adversarial supervision. Performance is rigorously quantified using mean Dice Similarity Coefficient (DSC), Intersection over Union (IoU), and 95th-percentile Hausdorff Distance (HD95) across four segmentation targets: background, Right Ventricle (RV), Myocardium (MYO), and Left Ventricle (LV). The results are striking: AttentionUNet trained with Generative Adversarial Network (GAN) adversarial supervision and Quantum-Behaved Particle Swarm Optimization (QPSO) establishes a new performance ceiling with Dice Similarity Coefficient of 0.8625, Intersection over Union of 0.7612, and 95th-percentile Hausdorff Distance of 1.90 mm -- approaching the inter-observer agreement of expert cardiologists. Equally remarkable, the proposed Quantum Convolutional Neural Network (QCNN) delivers a competitive Dice Similarity Coefficient of 0.7596 using only 1.95 million parameters -- sixteen times fewer than the standard UNet -- redefining the parameter efficiency frontier for cardiac segmentation. Across the majority of backbone architectures evaluated, Quantum-Behaved Particle Swarm Optimization consistently outperforms both classical Particle Swarm Optimization and Adaptive Moment Estimation, establishing quantum-inspired hyperparameter optimization as a practically viable, computationally non-burdensome strategy for deep medical image segmentation. These findings provide the field with its most comprehensive cross-architecture, cross-optimizer reference to date and deliver clear, actionable guidance for designing cardiac segmentation pipelines across diverse clinical deployment scenarios.