A Novel Framework using Intuitionistic Fuzzy Logic with U-Net and U-Net++ Architecture: A case Study of MRI Bain Image Segmentation

A research team has published a novel AI framework that significantly improves the accuracy of segmenting brain structures from MRI scans. The work, led by Hanuman Verma, Kiho Im, and Akshansh Gupta, addresses a core limitation in current deep learning models like U-Net and U-Net++, which struggle with the inherent uncertainty and vagueness in medical images. Their solution, named IFS U-Net and IFS U-Net++, integrates intuitionistic fuzzy logic directly into the model architecture. This allows the AI to process input data not as crisp, definite pixels, but as representations that account for doubt, effectively managing ambiguity caused by issues like the partial volume effect where tissues blend together.

The proposed models were rigorously evaluated on two major public MRI brain datasets: the Internet Brain Segmentation Repository (IBSR) and the Open Access Series of Imaging Studies (OASIS). Performance was measured using standard metrics including Accuracy, Dice Coefficient, and Intersection over Union (IoU). The results demonstrated that the intuitionistic fuzzy logic enhancement consistently boosted segmentation performance across these benchmarks. By more accurately delineating brain tissue boundaries and handling imprecise data, this framework represents a meaningful step forward in automated medical image analysis, potentially leading to more reliable tools for clinicians diagnosing conditions like tumors, atrophy, or other neurological disorders.