Exploring the Role of Deep Learning in Liver Segmentation

Delineating a target organ is the procedure known as biomedical segmentation. Liver cancer is the second most common cancer to cause death. The automatic liver segmentation from CT scans is a necessary job for the diagnosis of liver cancer disease. However, because of its irregular form and proximity to other organs, the liver is one of the more challenging organs to segment. For liver segmentation in the current research, datasets with only liver labels are typically used, the model is not resistant to cases involving multiple phases and vendors. AbdomenCT1K is a large and various abdominal CT datasets from 12 medical institutions, including cases with multiple diseases, vendors, and phases. To segment the liver, we use the Dense-UNet trained with AbdomenCT-1K, and it achieved the dice of 0.9271 and NSD of 0.9685. We compare different backbone networks in this research, and the results show that our model is more reliable and performs better than many other deep learning models in the prior study.