Comparative study on the accuracies of customized and universal models for organs-at-risk segmentation in cervical cancer(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2024年第11期
- Page:
- 1337-1342
- Research Field:
- 医学放射物理
- Publishing date:
Info
- Title:
- Comparative study on the accuracies of customized and universal models for organs-at-risk segmentation in cervical cancer
- Author(s):
- LIU Xuanyu1; 2; CHEN Shuying1; 2; GUO Feibao1; 2; 3; 4; CHEN Yanbin1; 2; 4; HE Qing1; 2; 4; L?Wenlong1; 2; 3; 4; CHEN Qi5; ZHANG Yimeng5; WANG Shaobin5; 6; CAI Chuanshu1; 2; 3; 4
- 1. Department of Radiotherapy, Cancer Center, the First Affiliated Hospital of Fujian Medical University, Fuzhou 350005, China 2. National Regional Medical Center, Binhai Branch of the First Affiliated Hospital of Fujian Medical University, Fuzhou 350212, China 3. Key Laboratory of Radiation Biology of Fujian Higher Education Institutions, Fuzhou 350005, China 4. School of Medical Imaging, Fujian Medical University, Fuzhou 350122, China 5. MedMind Technology Co., Ltd., Beijing 100083, China 6. Tsinghua Medicine, Tsinghua University, Beijing 100084, China
- Keywords:
- Keywords: cervical cancer automatic segmentation RT-Mind customized model
- PACS:
- R318；R811.1
- DOI:
- DOI:10.3969/j.issn.1005-202X.2024.11.003
- Abstract:
- Abstract: Objective To compare and analyze the differences between customized models and commercial universal models in the segmentation of organs-at-risk in cervical cancer, and to investigate the feasibility of customized models. Methods A retrospective analysis was conducted on 270 cervical cancer patients. Senior clinicians manually delineated organs-at-risk, including the bladder, rectum, small intestine, pelvic bone marrow, femoral heads, and kidneys. The cases were randomly selected to develop customized models, with 202 cases allocated to the training set, 38 cases to the test set, and 30 cases to the validation set. The universal and customized models were used for segmentation on the test set, and the automatic segmentation results obtained by the two models were compared with manual segmentation results to assess the performance of the customized model. Results Both customized model and universal model had comparable DSC values to manual segmentation, demonstrating satisfactory delineation outcomes (DSC values ranging from 0.7 to 0.9). However, in terms of deviation of centroid and 95% Hausdorff distance, the customized model surpassed the universal model. Conclusion Compared with the universal model, the customized model offers superior accuracy in delineating the structures of organs-at-risk in cervical cancer. As the customized model is optimized based on specific datasets, it provides precise support for clinical decision-making and holds promising applications in the treatment of cervical cancer.
Last Update: 2024-11-26