A Deep Learning Approach for Contour Interpolation

Chenxi Zhao, Ye Duan, Deshan Yang

Abstract: Contour interpolation is an important tool to expedite manual segmentation of anatomical structures. The conventional shape-based interpolation (SBI) algorithm, based on distance map calculation then interpolation, often performs sub-optimally when the two adjacent to-be-interpolated manual contours differ dramatically, especially near the superior and inferior borders of organs and for the gastrointestinal structures. In this study, we present a deep learning solution to improve the robustness and accuracy for contour interpolation, especially for these historically difficult cases. The deep contour interpolation model achieved a dice score of 0.95±0.06 and a mean DTA value of 1.08±2.31 mm, averaged on 3167 testing cases of all 16 organs. In a comparison, the results by the conventional SBI method were 0.94±0.08 and 1.50±3.63mm, respectively. For the difficult cases, the dice score and DTA value were 0.91±0.09 and 1.71±2.25 mm by the deep model, compared to 0. 86± 0. 13 and 3.42±5.88 mm by the conventional SBI method. A student t-test was applied to confirm that the performance improvements were statistically significant (p< 0. 05) for all cases. It could be useful to expedite the tasks of manual segmentation of organs and structures in the medical images.