Pancreas Segmentation in Abdominal CT Scans using Inter-/Intra-Slice Contextual Information with a Cascade Neural Network

Abstract

Automatic pancreas segmentation with high precision in Computed Tomography (CT) images is a fundamental issue in both medical image analysis and computer-aided diagnosis (CAD). However, pancreas segmentation is challenging because of the high variability in location and anatomy of the organs, while occupying only a very small part of the entire abdominal CT scans. Due to the rapid development of the CAD system and the urgent need for clinical treatment, the pancreas image segmentation with high precision is demanded. In this paper, we propose a new approach for automatic pancreas segmentation of CT images using inter-/intra-slice contextual information with a cascade neural network. Fully convolutional neural networks (FCN) are used to extract intra-slice contextual information for pancreas segmentation. Recurrent neural networks (RNNs) is introduced to extract inter-slice contextual information. With the setting bounding boxes, the proposed method outperforms the state-of-the-arts with an average Dice Similarity Coefficient (DSC) of 87.72 for NIH dataset with 4-fold cross-validation.