Application of AccuContour Software for Designing Vertebral Organ at Risk Models of Pediatric Tumors

Background: The AccuContour software was used to outline the spine and lumbar spine of children with solid tumors separately to evaluate and compare the outline effect, which provided the basis for the subsequent establishment of automatic outline models of the cervical and thoracic spine and further improved the accuracy of vertebral body outline. Methods: : Forty-three children with solid tumors were selected as group A, from which 37 children with solid tumors with intact lumbar vertebrae in the localized CT images were selected as group B, and were respectively imported into the artificial intelligence outlining system for manual outlining. In group A, all vertebrae in the image were outlined, and in group B, only the lumbar vertebrae were outlined, and the outlined images were deep-learnt in the system to establish the automatic outlined model A and model B. Thirteen children with solid tumors were selected, and the same children's vertebrae were automatically outlined based on models A and B, and manually outlined at the same time. The volume difference, position difference and shape consistency of the vertebral bodies were evaluated and compared between the automatic and manual outlining. Results: : Statistical analysis of the data from Model A and B groups showed statistically significant differences in ΔV%, HD95, and DSC （Dice Similarity Coefficient, DSC）values (P<0.05). Conclusion: The artificial intelligence automatic outlining system has a high accuracy in outlining vertebrae, and the accuracy of outlining lumbar vertebrae alone is higher than that of outlining all vertebrae within the image, and the automatic sketching models of cervical and thoracic vertebrae can be established subsequently to further improve the accuracy of vertebrae sketching.