Interactive Segmentation of Pancreases in Abdominal Computed Tomography Images and Its Evaluation Based on Segmentation Accuracy and Interaction Costs

The present paper proposed an interactive segmentation method of pancreases in abdominal computed tomography (CT) images based on the anatomical knowledge of medical doctors and the statistical information of pancreas shapes. This segmentation method consisted of two phases: training and testing. In the training phase, pancreas regions were manually extracted from sample CT images for training, and then a probabilistic atlas (PA) was constructed from the extracted regions. In the testing phase, a medical doctor selected seed voxels for a pancreas and background in a CT image for testing by use of our graphical user interface system. The homography transformation was used to fit the PA to the seeds. The graph cut technique whose data term was weighted by the transformed PA was applied to the test image. The seed selection, the atlas transformation, and the graph cut were executed iteratively. This doctor-in-the-loop segmentation method was applied to actual abdominal CT images of fifteen cases. The experimental results demonstrated that the proposed method was more accurate and effective than the conventional graph cut.