Interclass Balance Factor-Based Membership Fusion Semi-Supervised Fuzzy Clustering Algorithm for Lesion Segmentation in Cerebral Infarction Images

This study addresses the challenge of improving the accuracy and efficiency of lesion segmentation in brain stroke Diffusion Weighted Images (DWI), with a particular focus on the issue of class imbalance. Traditional clustering algorithms often fail to effectively segment brain stroke lesions due to the significantly smaller area of lesion regions compared to other brain tissues and background areas. To overcome this, we propose an Interclass Balance Factor-based Membership Fusion Semi-supervised Fuzzy Clustering Algorithm (ICBF-MFSFCM). This novel algorithm introduces an interclass balance factor to enhance the precision and consistency of clustering outcomes by better representing minority classes. The method was validated on actual brain DWI image datasets, demonstrating its superiority and reliability in improving lesion segmentation accuracy. The experimental results show that ICBF-MFSFCM outperforms traditional clustering algorithms in terms of Dice Coefficient (DSC), Intersection over Union (IoU), F1-score, and Surface Dice Similarity Coefficient (SDSC). These improvements offer a more efficient approach for the preliminary detection and treatment of cerebral stroke, contributing to better clinical outcomes for patients.