Click here to


Are you sure ?

Yes, do it No, cancel

Consensus Contouring Software with Real-Time Feedback for Radiation Oncology Training

J Yang1*, R Fang2 , L Court1 , Z Liao1 , D Gomez1 , J Gunther1 , J Yang1 , (1) UT MD Anderson Cancer Center, Houston, TX, (2) Rice University, Houston, TX


(Sunday, 7/29/2018) 3:00 PM - 6:00 PM

Room: Exhibit Hall

Purpose: To develop a software tool that can give real-time quantitative feedback for medical dosimetrists, medical residents and students in contouring training and guide them to improve their contouring consistency for radiation treatment planning.

Methods: We defined a new quantitative metric that includes spatial information, termed localized signed surface distance (LSSD), to give real-time feedback to trainees for contouring improvement at specific locations. The LSSD, represented as a color map, shows whether a trainee contour is under- or over-contouring in each small portion in a slice for each slice. We collected eight different expert contours on heart and left ventricle for 6 patients as initial training cases. More structures and training cases can be added in the future. After each trainee completes contouring a structure, a reference contour is generated by applying the simultaneous truth and performance level estimation algorithm to the eight expert contours and the trainee contour. The trainee contour is compared with the reference contour to create the LSSD color map. In this way, the LSSD takes into account the inter-observer variability in contouring assessment. We created a software tool with a friendly graphical user interface (GUI) using the LSSD color map for contouring training.

Results: To test the training interface a manually drawn heart contour was used to create the LSSD color map. The color map correctly identified regions of over- and under-contour. When the user clicked on the color map, the GUI correctly displayed the corresponding axial slice, and highlighted the portion of the contour that needs attention. This testing was repeated with different contours, and the GUI was found to be robust to different contouring errors.

Conclusion: We developed a software tool that can be used for consensus contouring training. The next step is to test the tool with a larger group of trainees.

Funding Support, Disclosures, and Conflict of Interest: This research was supported in part by The University of Texas MD Anderson Cancer Center Institutional Research Grant (IRG) Program.


Image Processing, Software, Segmentation


IM/TH- Image Analysis (Single modality or Multi-modality): Image segmentation

Contact Email