An Artificial Intelligence Approach for Automatic Delineation of Para-Aortic Clinical Target Volume

Abstract

Treatment of metastatic para-aortic lymph nodes (PAN) is critical for locally advanced cervical cancer (LACC) control. However, the clinical target volume (CTV) delineation for the para-aortic region is tedious and time-consuming even for a well-trained radiation oncologist. In this study, we trained a deep learning (DL) segmentation network that led to automatic and accurate PAN-CTV delineation.We retrospectively collected 47 radiotherapy planning CTs of LACC patients. For each planning CT, the Aorta, Inferior Vena Cava (IVC) and PAN-CTV were manually delineated for radiotherapy treatment planning for the elective para-aortic region from the left renal vein to the aortic bifurcation. The PAN-CTV was prophylactically contoured (ignoring gross para-aortic nodes) according to the recent Small et al. (2020) consensus guidelines. For each CT set, the spine structures were automatically delineated using threshold- and morphology-based segmentation. Then, all CTs were registered to a common reference frame based on the delineated spine structures and cropped to dimensions of 15 cm x 15 cm x 20 cm centered along the spine. A U-Net-based DL segmentation network was developed and trained to automatically segment Spine, Aorta, IVC, and PAN-CTV in the registered and cropped region. The segmented structures were converted into DICOM RTStruct with the original CT association.We used 32 CTs for training, 8 for validation, and 7 for testing. The Dice coefficient (DC) and Hausdorff distance (HD95) were used as the evaluation metrics. The mean (std) DC of PAN-CTV for training, validation and test sets were 0.911 (0.027), 0.757 (0.046) and 0.755 (0.073), respectively. The mean (std) HD95 of PAN-CTV for training, validation and test sets were 2.39 (2.77) mm, 10.12 (6.83) mm and 8.42 (4.52) mm, respectively. All testing PAN-CTVs were visually checked by an attending radiation oncologist specializing in gynecologic malignancies and deemed clinically acceptable.An Artificial Intelligence approach comprised of registration and DL network was developed and validated for automatic segmentation of Spine, Aorta, IVC, and PAN-CTV in LACC. Registration was used to improve the boundary definition of the virtual structure, PAN-CTV, for DL segmentation without human intervention. This approach could facilitate efficient workflow and allow for optimal dose delivery to the involved nodal basins in LACC.W. Chi: None. P. D'Cunha: None. M. Chen: None. L. Ma: None. M. Kazemimoghadam: None. Z. Yang: None. X. Gu: None. K.V. Albuquerque: Research Grant; Astra Zeneca. Honoraria; ACR, ARRT. Travel Expenses; ACR, ARRT, ASCO.W. Lu: None.