Deep Learning Automates Peritoneal Cancer Index Segmentation on CT

A novel deep learning method has been suggested by researchers to facilitate the automatic segmentation of radiological Peritoneal Cancer Index (rPCI) areas in CT scans, filling the gap left by the absence of a standardized imaging equivalent to the invasive surgical PCI (sPCI). The investigation tested nnU-Net and Swin UNETR on 62 CT scans, with rPCI areas manually marked by three clinical researchers and confirmed by two expert radiologists. The evaluation utilized five-fold cross-validation, measuring performance through the Dice Similarity Coefficient, 95th percentile Hausdorff distance, and Average Surface Distance. nnU-Net recorded an overall Dice score of 0.82. The rPCI regions are derived from a recent consensus study that outlines 3D anatomical areas for imaging assessment, while sPCI categorizes the abdomen into 13 regions assessed by tumor size during diagnostic laparoscopy.