Q-rung orthopair fuzzy framework for multimodal MRI image fusion and U-Net-based segmentation for brain tumor...

Medical image analysis is a significant research domain in contemporary healthcare systems, which helps to accurately diagnose diseases and support clinical decision-making. This domain consists of important operations, including image enhancement, fusion, segmentation, morphological processing, and denoising, which assist in the effective understanding of complex biomedical information. Multimodal image fusion and segmentation are particularly important among these processes, as they enable the integration of complementary diagnostic information and the clear interpretation of clinically meaningful structures. However, medical images often consist of uncertainty, noise, and incomplete pixel information, creating major challenges for conventional analytical techniques.

To address these challenges, the present study introduces a q-rung orthopair fuzzy set (qROFS)-based framework for multimodal magnetic resonance imaging (MRI) fusion, subsequently a U-Net-based segmentation architecture for brain tumor detection. To effectively handle uncertainty and preserve important image information, the fusion strategy is designed based on q-rung orthopair fuzzy Hamacher t-norm and t-conorm operators to integrate complementary information in four different MRI modalities, including T1, T1ce, T2, and Flair. The resultant fused images are subsequently processed through a U-Net architecture to obtain precise tumor segmentation.

Experiments conducted on the BraTS 2020 benchmark dataset demonstrate the effectiveness of the proposed framework. Across the four MRI modalities (T1, T1ce, T2, and Flair), the proposed fusion framework attains the best quantitative performance when compared to state-of-the-art methods with an SSIM of 0.9271, an MSSIM of 0.9481, a PSNR of 24.7130 dB, a low AMBE of 6.6779, a Correlation coefficient of 0.9583, an MI of 1.4097, and a UQI of 0.9767. A comparative evaluation with existing fuzzy set-based fusion models further assures the effectiveness of the proposed qROFS-based framework for multimodal MRI image fusion. Furthermore, the U-Net-based segmentation model achieves excellent performance, with an accuracy of 98.96%, a precision of 98.95%, a sensitivity of 98.94%, and a specificity of 99.65%.