Automatic liver tumor segmentation in abdominal CT scans using Convolutional LSTM-added U-net

Abstract

Liver plays an important role in metabolism and is a very important organ for us to maintain life. It is so necessary to manage the liver healthy but most of the diseases are usually manifested after it has progressed considerably. If it is possible to diagnose the disease at an early stage before delayed awareness, treatment effect and prognosis can be improved. We should pay more attention to the volume of internal lesions among several markers for early diagnosis of liver disease. There is a correlation between chronic liver diseases such as liver fibrosis or cirrhosis and the volume of liver, so it is meaningful to focus on it. In the case of interstitial tumors, its volume information can be utilized to determine the direction of future treatment by recognizing the progress at regular intervals. When trying to measure the volume of a specific organ from a 3D biomedical image, more accurate automatic segmentation method should be accompanied. There is still a demand for a method of automatically segmenting out overlooked tumors in the abdominal image of computed tomography and presenting it, so we propose CLU-net combining Convolutional LSTM for U-net and try to satisfy the demand in this paper. The conventional U-net applying the segmentation method for each 2D slice does not utilize the third axis information, but the CLU-net utilizes the information between the slices as a clue via the added Convolutional LSTM to demonstrate better performance.