Room: Exhibit Hall
Purpose: To determine the effect of increasing the atlas sizes of reference images on the quality and accuracy of autocontouring using ABAS in cervical cancer.
Methods: Four sets of atlases which cases of cervical cancer were 30, 60, 90 and 120 were established in MIM-Maestro (version 6.6.5). One experienced radiation oncologist contoured 4 OARs based on published consensus guidelines on 10 cervical cancer patients who were randomly selected out of the atlases, which were defined as reference volumes, Vref. 4 OARs on these 10 patients were auto-contoured respectively using ABAS, which were defined as Vac-30, Vac-60, Vac-90, Vac-120.3 and 9 were selected as the matching numbers of contouring. Using all available contours, the similarity of the ABAS and reference contours was assessed using dice similarity coefficient(DSC), sensitivity index(SI), inclusiveness index(InI), deviation of centroid, jaccard coefficient(JC) and maximum hausdorff distance(HDmax).The time for manual and auto-contouring the OARs was also recorded.
Results: The mean time of four sets of ABAS with the matching number 3(M3) showed a significant reduction than that of manual contouring (p<0.01), and a same trend (p<0.01) with the matching number 9(M9). But there were no difference among the groups of ABAS. The clear improvement was found between Vac-30 and Vac-120 in SI value(p=0.018,p=0.01) with M3 for bladder and rectum, DSC(p=0.016), JC(p=0.013) and HDmax(p=0.042) values with M3 for rectum, HDmax(p=0.002) values with M9 for rectum. And the other results had no significant difference.
Conclusion: The contouring time had no significant changes with addition of more patients to atlas. The improved accuracy in ABAS for the rectum of cervical cancer was found with the increase of atlas sizes.
Funding Support, Disclosures, and Conflict of Interest: Chinese Natural Science Fund(No. 61601012) Natural Key R&D Program of China(No.2017YFC0112100)
Not Applicable / None Entered.
Not Applicable / None Entered.