Automated segmentation of the stomach, liver, GB, pancreas, spleen, rib, skin, and abdominal wall was performed on the portal phase CT image using a deep learning algorithm based on fine-tuned 3D U-Net. 3D U-Net is a specialized deep learning algorithm for biomedical image segmentation, and this algorithm used its own fine-tuned model with learning from radiologists-annotated clinical data. On the early arterial and the portal phase CT images, biomedical engineers used semi-automatic segmentation software (AVIEW, Coreline Soft, Seoul, Korea) and segmented upper abdominal vessels which were essentially needed for making a surgery-oriented 3-D model as follows: aorta, celiac artery, left and right gastric arteries, splenic artery, common hepatic artery, proper hepatic artery, left hepatic artery, right hepatic artery, aberrant hepatic artery if present, gastroduodenal artery, left and right gastroepiploic arteries, inferior vena cava, portal vein, splenic vein, left gastric vein, and left and right gastroepiploic veins. For 3-D reconstruction, 3-D masks of organs and vessels were obtained from this segmentation process and inspected by one radiologist with 19 years of experience in abdominal imaging.
Free full text: Click here