An Intuitionistic Fuzzy Logic Driven UNet architecture: Application to Brain Image segmentation

Accurate segmentation of MRI brain images is essential for image analysis, diagnosis of neuro-logical disorders and medical image computing. In the deep learning approach, the convolutional neural networks (CNNs), especially UNet, are widely applied in medical image segmentation. However, it is difficult to deal with uncertainty due to the partial volume effect in brain images. To overcome this limitation, we propose an enhanced framework, named UNet with intuitionistic fuzzy logic (IF-UNet), which incorporates intuitionistic fuzzy logic into UNet. The model processes input data in terms of membership, nonmembership, and hesitation degrees, allowing it to better address tissue ambiguity resulting from partial volume effects and boundary uncertainties. The proposed architecture is evaluated on the Internet Brain Segmentation Repository (IBSR) dataset, and its performance is computed using accuracy, Dice coefficient, and intersection over union (IoU). Experimental results confirm that IF-UNet improves segmentation quality with handling uncertainty in brain images.

💡 Research Summary

The paper introduces IF‑UNet, a novel brain MRI segmentation framework that integrates intuitionistic fuzzy set (IFS) theory into the classic UNet architecture to address partial‑volume effects and boundary uncertainty. An IFS represents each pixel by three components: membership (µ), non‑membership (ν), and hesitation (π), where ν is derived using Sugeno’s negation function and π = 1 − µ − ν. The original MRI scans are transformed into a three‑channel fuzzy image, which is fed directly into the UNet encoder. Standard convolutional layers process all three channels simultaneously, allowing the network to learn features from both certainty (µ, ν) and uncertainty (π) information. Skip connections preserve these fuzzy features for the decoder, improving the reconstruction of ambiguous boundary regions. Experiments were conducted on the publicly available Internet Brain Segmentation Repository (IBSR) dataset, consisting of 20 T1‑weighted scans with expert annotations. Under identical training conditions, IF‑UNet was compared against a baseline UNet and an Attention‑UNet. Evaluation metrics—accuracy, Dice coefficient, and Intersection‑over‑Union (IoU)—showed consistent improvements for IF‑UNet (≈0.96 accuracy, 0.92 Dice, 0.86 IoU) over the baselines, indicating a 2‑5 % gain. Qualitative results demonstrated smoother segmentation contours, especially at tissue boundaries where hesitation values were high. The study acknowledges several limitations: the small sample size, lack of cross‑validation and statistical significance testing, and no quantitative analysis of the additional computational overhead introduced by fuzzy transformation. Future work is suggested to include larger multi‑center datasets, ablation studies on the individual contributions of µ, ν, and π, optimization of the λ parameter in Sugeno’s function, and extension to 3‑D volumetric segmentation. Overall, IF‑UNet offers a promising direction for incorporating uncertainty modeling into deep‑learning‑based medical image segmentation.