Effect of volume on the accuracy of organs-at-risk segmentation by automatic segmentation software(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2020年第7期
- Page:
- 797-802
- Research Field:
- 医学放射物理
- Publishing date:
Info
- Title:
- Effect of volume on the accuracy of organs-at-risk segmentation by automatic segmentation software
- Author(s):
- LI Hualing1; WANG Peipei2; LI Jinkai2; LI Caihong2; CHANG Zhigang2; GU Xiaohuan2; SUN Xinchen2; CAO Yuandong2
- 1. Department of Special Medicine, Nanjing Medical University, Nanjing 210009, China 2. Department of Radiation Oncology, the First Affiliated Hospital of Nanjing Medical University, Nanjing 210009, China
- Keywords:
- Keywords: AccuContour automatic segmentation software organs-at-risk volume geometric accuracy
- PACS:
- R811.1
- DOI:
- DOI:10.3969/j.issn.1005-202X.2020.07.001
- Abstract:
- Abstract: Objective To evaluate the geometric accuracy of artificial intelligence-based automatic segmentation system (AccuContour automatic segementation software) for the segmentation of organs-at-risk and to investigate whether the geometric accuracy of automatic OAR segmentation is affected by the volume of OAR. Methods A total of 161 patients were enrolled, including 40 cases with tumor in head and neck, chest, and pelvis, and 41 cases with tumor in the abdomen. AccuContour automatic segementation software and manual delineation were used to contour lens, optic nerve, eyeball, pituitary, brainstem, parotid glands, mandible, lungs, heart, kidneys, liver, rectum, bladder, femoral heads and other OAR. The Dice similarity coefficient (DSC), Hausdorff distance (HD), Jaccard coefficient and volume of each segmented OAR were analyzed, and the geometric accuracy of the automatic segmentation was evaluated using DSC, Jaccard coefficient and HD. Spearman correlation analysis was used to verify the correlations between volume and DSC, HD, Jaccard coefficient and Wilcoxon paired rank sum test was conducted on the DSC, HD and Jaccard coefficients of paired OAR such as lens, optic nerves, eyeballs, parotid glands, temporomandibular joints, lungs, kidneys and femoral heads, etc. Results The mean DSC of each OAR was larger than 0.7. The mean Jaccard coefficient of the neck was between 0.557 and 0.880, with the lowest Jaccard coefficient in the right optic nerve and the highest in the mandible. The mean HD of head and neck, chest, and pelvis was less than 8, 22 and 16 mm, respectively. Except for the liver (HD=34.563 mm), the other OAR in the abdomen had a HD less than 19 mm. There were correlations between volume and DSCall, Jaccardall coefficient, HDall (rDSC=0.757, PDSC=0.000 rJaccard=0.775 PJaccard=0.000 rHD=0.761, PHD=0.000). The DSC and Jaccard coefficient of lens, optic nerves, eyeballs, pituitary, brainstem, parotid glands, mandible, lungs, bladder, femoral heads, etc. were also correlated with the volume of OAR (P<0.05), but their HD was not related to volume (P>0.05). The differences in DSC and Jaccard coefficient between bilateral lungs were statistically significant (PDSC =0.000, PJaccard=0.000). Conclusion AccuContour automatic segmentation software has a high accuracy for OAR segmentation, and the geometric accuracy of automatic segmentation is affected by the volume of OAR.
Last Update: 2020-07-28