FEARE: A fuzzy entropy-driven adaptive region energy active contour model for medical image segmentation

Abstract

Image segmentation plays a critical role in medical image analysis, where intensity inhomogeneity, noise, and weak boundaries often degrade performance. Many region-based active contour models rely on fixed assumptions, limiting their adaptability in complex clinical scenarios. To address this limitation, we propose FEARE, a fuzzy entropy-driven adaptive region energy active contour model within a level-set framework. The model incorporates entropy-guided confidence weighting to dynamically adjust region fitting in uncertain regions, while a geometry-aware regularization term preserves anatomical plausibility during contour evolution. The proposed method is evaluated on the Sunnybrook Cardiac MRI dataset using overlap, boundary, and efficiency metrics. FEARE achieves a mean Dice Similarity Coefficient of 0.846 and an Intersection over Union of 0.757, outperforming several state-of-the-art active contour models with fewer segmentation failures, while maintaining competitive computational efficiency. These results demonstrate that entropy-driven adaptive modeling improves segmentation robustness without requiring training data. The FEARE framework provides an interpretable and data-efficient alternative to learning-based methods for medical image segmentation. The evaluation scripts and source code are publicly available to make them reproducible. The FEARE model's source code is available at https://doi.org/10.5281/zenodo.1555982, and the Sunnybrook Cardiac MRI dataset is available at https://www.cardiacatlas.org/sunnybrook-cardiac-data/.