MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Clinical Evaluation of Atlas and Deep Learning-Based Automatic Contouring of Multiple Organs at Risk and Clinical Target Volumes for Breast Cancer

M Choi1*, B Choi2, S Chung3, N Kim4, J Chun5, Y Kim6, J Chang7, J Kim8, (1) Department of Radiation Oncology, Yonsei Cancer Center, Seoul, KR, (2) Department of Radiation Oncology, Yonsei University College of Medicine, Seoul, KR, (3) Department of Radiation Oncology, Yonsei University College of Medicine, Seoul, KR, (4) Department of Radiation Oncology, Yonsei University College of Medicine, Seoul, KR, (5) Department of Radiation Oncology, Yonsei University College of Medicine, Seoul, KR, (6) Department of Radiation Oncology, Yonsei University College of Medicine, Seoul, KR, (7) Department of Radiation Oncology, Yonsei University College of Medicine, Seoul, KR, (8) Department of Radiation Oncology, Yonsei Cancer Center, Seoul, KR

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: The contouring of organs at risk (OARs) and clinical target volumes (CTVs) is an important aspect of radiotherapy planning. However, this task is time-consuming and prone to inter-observer variation. Given this issue, interest for auto-segmentation has been rising. The aim of our study was to determine the clinical feasibility of auto-segmentation methods for target and normal organs and specifically evaluate the feasibility of a deep-learning-based auto-segmentation (DLBAS) compared to the commercially released atlas-based auto-segmentation (ABAS) solutions.


Methods: Contrast-enhanced planning CT data from 54 patients with breast cancer who underwent breast-conservation surgery was used in this study. Contours of OARs, CTVs and heart sub-structures were generated by ABAS from MIM and Mirada with 35 atlas library subjects and DLBAS using a fully convolutional DenseNet (FCDN) with 35 training sets. The accuracy of segmentation was assessed on six test patients using the Dice coefficient with reference to the manually delineated contours.


Results: Compared to ABAS, the proposed FCDN model yielded more consistent results and the highest average DSC in the majority of the structures, especially the CTVs where FCDN produced 80%, MIM and Mirada produced 68% and 71%, respectively. In the OARs, FCDN produced an average DSC of 89% whereas MIM and Mirada produced 82% and 84%, respectively. Lastly, in the heart and its substructures, apart from the coronary arteries, the results of FCDN (86%) and ABAS (MIM:83%, Mirada:86%) were comparable.


Conclusion: We assessed the clinical feasibility of the ABAS of MIM and Mirada and DLBAS using FCDN algorithms for the segmentation of target volumes and OARs including heart substructures. Compared to the ABAS, DLBAS using the FCDN algorithm generated a more consistent and robust performance across most structures. As a preclinical study, we have confirmed the plausibility for clinical implementations of these segmentation solutions.

Keywords

Segmentation, Breast, Validation

Taxonomy

Not Applicable / None Entered.

Contact Email