MENU

Click here to

Ă—

Are you sure ?

Yes, do it No, cancel

Evaluation of Deep Learning-Based Auto-Segmentation of Target Volume and Normal Organs in Breast Cancer Patients

SY Chung1*, JS Chang1, Y Chang2, BS Choi1, J Chun1, JS Kim1, YB Kim1, (1) Department of Radiation Oncology, Yonsei University College of Medicine, Seoul, KR, (2) CorelineSoft, Co., Ltd, KR

Presentations

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

Room: AAPM ePoster Library

Purpose: Intensity-modulated radiotherapy (IMRT) allows lower radiation doses to the nearby normal organs, thus reducing toxicity in breast cancer patients. However, target volume and normal organ segmentation for IMRT increases the physicians’ workload considerably. Thus, deep learning-based auto-segmentation can be an expedient tool for target and normal organ segmentation. Here, we evaluated the deep learning-based auto-segmented contours compared to manually delineated contours in breast cancer patients.


Methods: Clinical target volumes (CTV) for bilateral breasts and lymph node and normal organs including heart, lung, esophagus, spinal cord, thyroid and cardiac substructures (atrium and ventricle, left ascending artery (LAD), right coronary artert (RCA)) were manually delineated on a planning computed tomography scans of 61 breast cancer patients. Afterwards, a two-stage convolutional neural network algorithm was conducted. Quantitative metrics, including dice similarity coefficient (DSC) and Hausdorff distance (HD), and qualitative scoring by expert and non-expert panel were used for analysis.


Results: The correlation between the auto-segmented and manual contours was excellent for CTV and normal organs except for LAD and RCA. Auto-segmented contours for LAD and RCA showed reduced performance with mean DSC lower than 0.5 and mean HD higher than 20 mm, whereas other CTV and normal organs showed mean DSC of mostly higher than 0.80. Qualitative subjective scoring showed good results for all CTV and normal organs.


Conclusion: The feasibility of deep learning-based auto-segmentation was shown in this study. Although deep learning-based auto-segmentation cannot be a substitution for radiation oncologists, it can be an expedient tool in clinic, by assisting radiation oncologists.

Funding Support, Disclosures, and Conflict of Interest: This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (2019R1C1C1009359).

Keywords

Segmentation

Taxonomy

IM/TH- image segmentation: CT

Contact Email