Resource-Aware Evolutionary Neural Architecture Search for Cardiac MRI Segmentation

2026-05-06

AI Summary

This work addresses the challenges of cardiac MRI segmentation caused by low contrast, ambiguous boundaries, and inter-scan variability by proposing CardiacNAS—a resource-aware evolutionary neural architecture search framework. The method defines a cardiac segmentation-oriented search space within a UNet-like supernet, innovatively incorporating attention mechanisms, feature fusion strategies, and residual scaling as searchable components. A multi-objective evolutionary algorithm jointly optimizes segmentation accuracy—measured by Dice similarity coefficient (DSC) and Hausdorff distance at 95% (HD95)—alongside model efficiency in terms of parameters and FLOPs. Evaluated on the ACDC dataset, CardiacNAS achieves an average DSC of 93.22% and HD95 of 4.73 mm with only 3.58 million parameters and 14.56 GFLOPs, significantly outperforming six state-of-the-art methods.

Abstract

Cardiac magnetic resonance (CMR) segmentation underpins quantitative assessment of ventricular structure and function, yet reliable delineation remains difficult due to low tissue contrast, fuzzy boundaries, and inter scan variability. We present CardiacNAS, an evolutionary neural architecture search (NAS) framework that couples a UNet like supernet with a cardiac aware search space spanning depth width, kernel size, filter size, attention, fusion, activation, dropout, and residual scaling. The search is explicitly resource aware, jointly optimizing dice similarity coefficient (DSC) and 95th percentile Hausdorff distance (HD95) versus model size and floating point operations (FLOPs) under fixed compute budgets. Candidate architectures are instantiated from the supernet, trained with proxy budgets, and evolved through crossover, mutation, and elitist selection. We evaluate on the ACDC dataset and compare against six state of the art methods, using qualitative comparisons, learning curve analyses, and design factor correlation studies. The resulting model attains 93.22% average DSC and 4.73 mm HD95 with 3.58M parameters and 14.56 GFLOPs, demonstrating a favorable accuracy efficiency trade off. Analyses indicate that searched attention and fusion choices, together with residual scaling, contribute to improved boundary fidelity and stability. CardiacNAS offers a principled, resource aware approach to deployable CMR segmentation with transparent reporting of architectural complexity and compute budgets.

Problem

• Cardiac MRI segmentation
• low tissue contrast
• fuzzy boundaries
• inter-scan variability
• resource-aware

Innovation

• Neural Architecture Search
• Resource-Aware Optimization
• Cardiac MRI Segmentation
• Evolutionary Algorithm
• Attention Mechanism