Intensity-modulated radiotherapy planning for breast cancer based on two-layer perceptron neural network(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2023年第6期
- Page:
- 661-666
- Research Field:
- 医学放射物理
- Publishing date:
Info
- Title:
- Intensity-modulated radiotherapy planning for breast cancer based on two-layer perceptron neural network
- Author(s):
- LIU Weikun; ZHOU Linghong
- School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China
- Keywords:
- Keywords: breast cancer intensity-modulated radiotherapy machine learning automatic planning perceptron dose prediction
- PACS:
- R318；R815
- DOI:
- DOI:10.3969/j.issn.1005-202X.2023.06.001
- Abstract:
- Abstract: Objective To establish a two-layer perceptron neural network based on the anatomical characteristics of the left-sided breast cancer target area, the planning parameters of the corresponding radiotherapy plan and some key dose-volume histogram (DVH) information for realizing the automatic radiotherapy planning and the prediction of some key DVH information. Methods The two-layer perceptron neural network was trained with the overlapping volume histogram features and tangential angles in the intensity-modulated radiotherapy (IMRT) plans of 50 cases of left-sided breast cancer as the input, and the field angles, objective function parameters and some key DVH information as output. The predicted field angles and objective function parameters for IMRT plans of 10 new cases were used for automatic planning. The dosimetric results of automatic plans and manual plans and some key DVH information predicted by the network were analyzed using paired t-test to verify the feasibility of the method for automatic IMRT planning for left-sided breast cancer and test the accuracy of prediction on some key DVH information. Results The automatic plans basically met the clinical requirements. The differences between automatic plans and manual plans in the Dmin, Dmax, Dmean, Vprescription, D5, D95, CI, HI of PTV, the V20, V5 of ipsilateral lung and the V20 of heart were trivial (P>0.05). The time required for automatic planning was much less than that required for manual planning. The predicted key DVH information included the Dmin, Dmax, Dmean, Vprescription of PTV, the V20, V5 of ipsilateral lung and the V20 of heart. Only the Dmean, Vprescription of PTV and the V20 of ipsilateral lung in automatic plans were different from those in manual plans, but all of them conformed to the needs and expectations of clinical radiotherapy. Conclusion The automatic planning with the two-layer perceptron neural network can result in the same plan quality as manual plans, and greatly shorten planning time. The predicted dosimetric information can also provide reference for the plan quality evaluation.
Last Update: 2023-06-28