Minkowski Distance-Driven FCM with PSO Optimization for Robust Segmentation of Brain Tumors in Medical Imaging

Abstract

Medical image segmentation plays a pivotal role in computer-aided diagnosis, particularly in detecting and analyzing brain tumors using Magnetic Resonance Imaging (MRI). Accurate segmentation is essential for effective treatment planning and prognosis. Traditional methods such as K-Means and Genetic Algorithm-based segmentation often struggle with intensity inhomogeneity, noise, and irregular tumor boundaries. Although the Minkowski distance-based segmentation improves accuracy by incorporating geometric adaptability, it still lacks the intelligence to handle complex and heterogeneous tumor structures effectively. To address these limitations, this paper proposes a novel Hybrid Minkowski–Driven Fuzzy C-Means (FCM) with Particle Swarm Optimization (PSO) and Deep Learning (HMDL) framework for robust brain tumor segmentation and classification. The Minkowski distance metric enhances adaptive clustering by capturing spatial similarity, while the FCM algorithm ensures precise boundary delineation through fuzzy membership modeling. Further, PSO optimization dynamically fine-tunes the clustering parameters to achieve optimal convergence and stability. The deep learning module, built upon a U-Net-based convolutional neural network, refines segmentation outputs and enables accurate classification of tumor regions.

In addition to traditional 2D slice-based processing, the proposed framework incorporates 3D volumetric data segmentation using multi-modal MRI (T1, T2, FLAIR) to ensure inter-slice spatial consistency and precise volumetric tumor representation. This 3D integration significantly enhances the model’s ability to capture anatomical continuity across slices, leading to more reliable and clinically relevant segmentation outcomes. Experimental evaluations on the BRATS 2021 dataset demonstrate that the proposed HMDL model significantly outperforms existing conventional and hybrid segmentation techniques in terms of Dice coefficient, Intersection over Union (IoU), Hausdorff distance, accuracy, and computational efficiency. This study introduces a comprehensive, intelligent, and interpretable hybrid framework that successfully combines mathematical distance metrics, optimization algorithms, 3D data analysis, and deep learning to achieve superior performance in medical image analysis and brain tumor detection.