Room: AAPM ePoster Library
Automated segmentation of organs at risk (OARs) is usually not as accurate as that contoured by experts. However, the primary concern of radiotherapy is not the accuracy but the dosimetric impacts of the contours. Here, we proposed a plan quality-driven method to evaluate automated segmentation for radiotherapy.
The patient data included 19 cases with esophagus cancer. The physicians manually contoured the target volumes and the OARs. A state-of-the-art convolutional neural networks (CNN) was used to contour all the OARs automatically. The OARs included spinal cord PRV, heart, left lung and right lung. Two sets of radiotherapy plans with manually or automated contoured OARs were generated using the same beam configuration and optimizing parameters, respectively. Then the dosimetric metrics of the plans were calculated with the manual contours.
The segmentation accuracy for spinal cord PRV, heart, left lung, right lung, and lungs were as follows: Dice similarity coefficient: 0.92±0.02, 0.93±0.04, 0.96±0.01, and 0.95±0.02; mean distance to agreement (mm): 0.74±0.22, 1.74±0.85, 1.07±0.27, 1.63±0.43, and 1.32±0.26, respectively. All the plans met the dosimetric requirement. The following metrics of the targets and parallel OARs were comparable (p>0.05): homogeneity index (HI) and conformity index (CI) of the target volume; mean dose, V10, V20, and V30 of left lung, right lung, and lungs; and mean dose, V30, and V40 of heart. The only metric having significantly statistical difference was the maximum dose of the spinal cord PRV (p=0.035); however, this dosimetric parameter had no relevance to the segmentation accuracy.
The findings demonstrated that all the parallel OARs can be directly replaced by automated segmentations for the radiotherapy planning of esophagus cancer. But for the serial organ, of which the important dosimetric metric is the maximum dose, careful human review will be required.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by the Beijing Hope Run Special Fund of Cancer Foundation of China (LC2019B06, LC2018A14), the Beijing Municipal Science & Technology Commission (Z181100001918002), and the National Natural Science Foundation of China (11975313).