Deep learning of 3D Computed Tomography (CT) images for organ segmentation using 2D multi-channel SegNet model

摘要

Purpose To accurately segment organs from 3D CT image volumes using a 2D, multi-channel SegNet modelconsisting of a deep Convolutional Neural Network (CNN) encoder-decoder architecture.Method We trained a SegNet model on the extended cardiac-torso (XCAT) dataset, which was previouslyconstructed based on patient Chest–Abdomen–Pelvis (CAP) Computed Tomography (CT) studies from 50 Dukepatients. Each study consists of one low-resolution (5-mm section thickness) 3D CT image volume and itscorresponding 3D, manually labeled volume. To improve modeling on such small sample size regime, we performedmedian frequency class balancing weighting in the loss function of the SegNet, data normalization adjusting forintensity coverage of CT volumes, data transformation to harmonize voxel resolution, CT section extrapolation tovirtually increase the number of transverse sections available as inputs to the 2D multi-channel model, and dataaugmentation to simulate mildly rotated volumes. To assess model performance, we calculated Dice coefficients ona held-out test set, as well as qualitative evaluation of segmentation on high-resolution CTs. Further, weincorporated 50 patients high-resolution CTs with manually-labeled kidney segmentation masks for the purposeof quantitatively evaluating the performance of our XCAT trained segmentation model. The entire study wasconducted from raw, identifiable data within the Duke Protected Analytics Computing Environment (PACE).Result We achieved median Dice coefficients over 0.8 for most organs and structures on XCAT test instances andobserved good performance on additional images without manual segmentation labels, qualitatively evaluated byDuke Radiology experts. Moreover, we achieved 0.89 median Dice Coefficients for kidneys on high-resolution CTs.Conclusion 2D, multi-channel models like SegNet are effective for organ segmentations of 3D CT image volumes,achieving high segmentation accuracies.