An Efficient Prostate MRI Segmentation using Deep Learning for better Cancer Diagnosis

Prostate cancer is a prevalent malignancy in men, often diagnosed using magnetic resonance imaging (MRI). Accurate segmentation of prostate MRI is crucial for early cancer diagnosis, yet the challenges arise from limited annotated data and variations in prostate shape, appearance, and size. This study introduces an innovative U-shaped approach called UMC-Net that combines an encoder of MaxVIT and a novel convolution block, enabling the fusion of global and local features to enhance prostate MRI segmentation. The Atrous convolution extracts spatial dimension information, followed by point-wise convolution, which reduces parameters and computations. The Global and local features fusion (GLFF) module performs a better fusion of features to improve the segmentation performance. The performance of the proposed method is evaluated on a publicly available dataset using the dice similarity coefficient (DSC) and Hausdorff distance (HD), which demonstrate remarkable results, with DSC and HD values of 0.887 and 0.761, 4.8, and 9.6, respectively. These findings highlight the superior performance of UMC-Net compared to the state-of-the-art methods in prostate MRI segmentation, paving the way for more accurate early cancer diagnosis.