ESTRO 2024 - Abstract Book

S983

Clinical - Gynaecology

ESTRO 2024

Clinical Practice of AI-Assisted Radiotherapy with Low-Dose FBCT-guided OART for Cervical Cancer

Chen-ying Ma 1 , Jian Guo 1 , Hua Jiang 1 , Zhi-bin Li 1 , Shi-mei Wang 2 , Bing-huan Li 3 , Le-cheng Jia 4 , Ran Cao 3 , Wei Zhang 3 , Xiao-ting Xu 1 , Ju-ying Zhou 1 1 the First Affiliated Hospital of Soochow University, Radiation Oncology, Suzhou, China. 2 Shanghai United Imaging Healthcare Co., Ltd, Academia Sinica, Shanghai, China. 3 Shanghai United Imaging Healthcare Co., Ltd, Radiotherapy department, Shanghai, China. 4 Shenzhen United Imaging Research Institute of Innovative Medical Equipment, Radiotherapy laboratory, Shenzhen, China

Purpose/Objective:

Online Adaptive Radiation Therapy (OART) has shown good efficacy in postoperative adjuvant radiotherapy and radical radiotherapy for cervical cancer. This study explored the feasibility and safety of artificial intelligence (AI)- assisted combined low-dose fan beam computed tomography (FBCT)-guided OART in the treatment of cervical cancer.

Material/Methods:

On the United Imaging uCT-ART platform, 11 patients with cervical cancer (10 postoperative adjuvant, 1 radical radiotherapy) underwent radiation oncologist (RO)-led trigger OART. Analysis of the feasibility of human AI assisted low-dose FBCT-guided OART in the treatment of cervical cancer, including evaluating the quality of automatic segmentation contours, automatic radiotherapy planning, OART online dosimetry analysis, OART process time and analysis of radiotherapy-related toxicity in 11 cervical cancer patients. This study was approved by the Chinese Clinical Trial Registry (No: ChiCTR2300074980).

Results:

In 297 fractional treatments, a total of 81 OARTs were initiated by RO, with an average of 7.4 OARTs per capita. The average total time of the OART process was 18 minutes and 58 seconds, and the average time of OART adjustment was 15 minutes and 52 seconds. In 11 paatients, the region of interest (ROI) was automatically segmented by the AI-assisted delineation tool on the positioning CT, and ROIauto was modified and reviewed by RO to obtain ROIedit, in which the dice similarity coefficient of the clinical target volume (CTV) was 0.85 ± 0.04, which was not inferior to the previous model 0.89±0.02 (P > 0.05), and the 95% Hausdorff distance was 5.64 ± 1.60 mm, which was better than the previous model construction 6.28±2.31 mm (P < 0.05). The outline delineation strategy after OART was started with the CTV rigid copy/OAR automatic segmentation + RO modification. The PTV dose distribution of the OART plan was more compact, and the overall dose was closer to the prescribed dose; the OAR dose control was closer to the clinical requirements, and the OAR obtains stricter prescription dose protection; the conformity index and uniformity index were better than the manual plan. The 2D-γ average pass rates of all beams under the 3%/3mm benchmark were 99.1%, and the minimum pass rates were 97.3%. The OART plan has passed the quality control requirements. During the course of OART, the variation range of target volume between fractions was mainly concentrated within the range of plus or minus 5%, while the variation range of OAR volume fluctuated greatly without obvious rules. The acute hematopoietic system toxicity in 11 patients at the end of OART was mainly decreased white blood cell count, neutrophil count and anemia, and the gastrointestinal system toxicity was mainly diarrhea symptoms. Acute toxic reactions were all CTCAE grade 1-2, and no toxic reactions of grade 3 or above occurred.

Conclusion: