A Neutrosophic Set‐Based Active Contour Model for Medical Image Segmentation

ABSTRACT This paper presents a novel neutrosophic set‐based active contour model for accurate and efficient segmentation of magnetic resonance (MR) images. A new level set formulation integrates neutrosophic truth, falsity, and indeterminacy subsets with the Heaviside function, replacing conventional gradient‐based information to enhance robustness against noise, intensity inhomogeneity, and weak or discontinuous edges. Adaptive α ‐mean and β ‐enhancement parameters, computed from image entropy, regulate local uncertainty, while an optimised averaging window ensures accurate boundary localisation without excessive smoothing. The proposed framework applies an adaptive contour updating strategy to improve convergence stability and reduce computational cost. Experimental validation on representative datasets demonstrates that the model achieves high segmentation performance (FOM up to 0.904, PRI up to 0.927) with superior efficiency, requiring as few as two iterations and 0.12 s runtime. Comparative analyses with ORACM and a U‐Net baseline highlight the robustness of the proposed method, particularly for images with limited annotations or challenging intensity variations. Failure cases and methodological limitations are discussed to provide balanced insight and guide future improvements. Overall, the proposed approach offers a mathematically interpretable, data‐efficient, and computationally fast alternative to conventional active contour and deep learning‐based segmentation methods.