Application of artificial intelligence and automated scripts in3D printing brachytherapy
10.13491/j.issn.1004-714X.2025.03.019
- VernacularTitle:探索在三维打印近距离治疗中应用人工智能及自动化脚本
- Author:
Wentai LI
1
;
Jiandong ZHANG
1
;
Zhihe WANG
2
;
Xiaozhen QI
3
;
Yan DING
1
;
Baile ZHANG
1
;
Wenjun MA
1
;
Yao ZHAI
1
;
Weiwei ZHOU
1
;
Yanan SUN
1
;
Xin ZHANG
1
Author Information
1. The First People’s Hospital of Nanyang City, Nanyang 473000, China.
2. The Second People's Hospital of Nanyang City, Nanyang 473000, China.
3. The Central Hospital of Luohe City, Luohe 462000, China.
- Publication Type:OriginalArticles
- Keywords:
Artificial intelligence;
Neural network;
Three-dimensional printing;
Brachytherapy;
Transformer;
U-Net;
Python automated script
- From:
Chinese Journal of Radiological Health
2025;34(3):419-425
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore the efficiency improvement in segmenting neural network with the application of Transformer + U-Net artificial intelligence (AI) and modeling with the application of Python scripts in three-dimensional (3D) printing brachytherapy. Methods A Transformer + U-Net AI neural network model was constructed, and Adam optimizer was used to ensure rapid gradient descent. Computed tomography or magnetic resonance imaging data of patients were standardized and processed as self-made data sets. The training set was used to train AI and the optimal result weight parameters were saved. The test set was used to evaluate the AI ability. Python programming language was used to write an automated script to obtain the output segmentation image and convert it to the STL file for import. The source applicator and needle could be automatically modeled. The time of automatic segmentation and modeling and the time of manual segmentation and modeling were entered by two people, and the difference was verified by paired t-test. Results Dice similarity coefficient (DSC), mean intersection over union (MIOU), and Hausdorff distance (HD95) were used for evaluation. DSC was 0.9341, MIOU was 0.8762, and mean HD95 was 2.516. The mean time of manual segmentation was 1187 s, and the mean time of AI segmentation was 145 s (P < 0.01). The mean time of manual modeling was 321 s, and the mean time of modeling with Python automated scripts was 18 s (P < 0.01). Conclusion The application of Transformer + U-Net automatic segmentation and Python automated scripts can effectively improve work efficiency in the 3D printing brachytherapy for cervical cancer.
