NEURONOVA-NET: ENTROPY-GUIDED DEEP LEARNING FRAMEWORK FOR NOISE REDUCTION AND BIAS CORRECTION IN MRI BRAIN IMAGES

Abstract

Magnetic Resonance Imaging (MRI) is of vital importance for the diagnosis of brain disease and clinical decision-making. Nonetheless, MRI scans are typically plagued by noise, bias field distortion, motion artifact, and intensity inhomogeneity, which affect diagnostic accuracy. To counter these limitations, this paper puts forward NeuroNova-Net (Neural Novel Enhancement and Refinement Network), a cutting-edge hybrid architecture that combines mathematical modeling with deep neural learning for complete MRI preprocessing. The proposed algorithm includes several stages—noise estimation and removal, correction for bias field, intensity standardization, artifact removal, and contrast enhancement—and provides improved image uniformity and anatomical integrity. NeuroNova-Net leverages anisotropic diffusion-based smoothing with CNN-induced residual learning and entropy-controlled artifact removal to produce best possible balance between denoising and structural information preservation. Experimental testing on publicly released datasets like BraTS 2021 and IXI proves that NeuroNova-Net surpasses state-of-the-art preprocessing models like ADF, N4-HM, and GAN-ASN with a mean PSNR of 33.45 dB and a minimum RMSE of 0.0465. The quantitative and qualitative results prove that NeuroNova-Net produces high-quality, bias-corrected, and artifact-free MRI images acceptable for downstream neuroimaging tasks like segmentation, classification, and lesion analysis.