[2507.15193] A Study of Anatomical Priors for Deep Learning-Based Segmentation of Pheochromocytoma in Abdominal CT

View the PDF file for the paper entitled Study of the Perceptions of the Deep Retail based on learning from the tumor in the abdomen, written by Tanjin Taer Toma and 13 other authors

PDF HTML (experimental) view

a summary:The exact partition of the PCC in the computerized tomography tests in the abdomen is necessary to estimate the burden of the tumor, diagnosis and planning for treatment. It may also help in concluding genetic groups, which reduces dependence on expensive tests. This study systematically evaluates the anatomical protein to determine the configurations that improve the learning -based PCC fragmentation. We used the NNU-Net framework to evaluate the eleven with the eleven explanation strategies for the three-dimensional retail for the tumor of the forces, as it provided a set of multi-layer layers based on the organic organizations of the organs. These fares were derived from the neighboring organs that usually surround adrenal gland tumors (for example, liver, spleen, kidney, aortic artery, adrenal gland, and pancreas), and were compared with the broad body area used in previous work. The frame was trained and tested on 105 improved CT scan tests on the variation of 91 patients at the Clinical Center of the National Institutes of Health. Performance was measured using the similarity of dice (DSC), the normal surface (NSD), and the exciting F1 degree. Among all strategies, the tumor explanation + kidney + aorta (TKA) has achieved the highest accuracy of division, greatly outperforming the tumor + the previously used body (TB) via DSC (P = 0.0097), NSD (P = 0.0110), and F1 degrees (25.84 % of Theresold of IOU from 0.5). The TKA model also showed the quantum measurement of tumor burdens (R^2 = 0.968) and strong retail across all genetic sub -species. In checking the health of five times, TKA has constantly outperformed TB via IOU sills (0.1 to 0.5), which enhances durability and generalization. These results highlight the value of the relevant anatomical context integrating into deep learning models to achieve a precise PCC segment, providing a valuable tool to support clinical evaluation and monitoring longitudinal diseases in PCC patients.