Automated Visceral Adipose Tissue Segmentation and Quantification from Abdominal MRI using an Enhanced U-Net and Region-growing

The study focuses on enhancing the accuracy and reliability of visceral adipose tissue (VAT) segmentation and quantification from abdominal MRI images. Accurate segmentation of VAT is crucial for assessing obesity-related health risks, as traditional methods struggle with irregular shapes and varying intensities. The research utilizes a methodology consisting of three key modules: homomorphic filtering for intensity inhomogeneity correction, a U-Net architecture with attention mechanisms for primary segmentation, and a region-growing algorithm for refining segmentation. Homomorphic filtering effectively separates bias fields, enhancing image quality by transforming multiplicative artifacts into additive ones and removing them with high-pass filtering. This process ensures precise segmentation by maintaining high-frequency anatomical details. The U-Net model incorporates attention mechanisms and skip connections to focus on VAT regions, utilizing both local and global image contexts.The Combined Healthy Abdominal Organ Segmentation (CHAOS) challenge dataset and the Cancer Imaging Archive (TCIA) dataset are used to train and evaluate the model. It achieves a Dice Similarity Coefficient (DSC) of up to 0.985 on the CHAOS dataset and 0.972 on the TCIA dataset, outperforming existing methods in terms of segmentation accuracy. The region-growing algorithm further refines the segmentation by expanding VAT regions from high-confidence seed points, ensuring accurate boundary delineation and reducing noise. The study's results, evaluated using k-fold cross-validation, show that the proposed methodology significantly improves VAT segmentation efficiency, achieving a median DSC of 0.96 for the CHAOS dataset and 0.95 for the TCIA dataset in the most comprehensive experimental scenario. Comparative analysis indicates that the proposed approach outperforms other models, with higher sensitivity and specificity values, highlighting its potential for clinical applications in obesity management..