Feasibility study of automatic assessment of abdominal and pelvic CT radiation dose based on deep learning algorithm
10.3760/cma.j.cn112271-20231024-00136
- VernacularTitle:基于深度学习算法的腹盆部CT辐射剂量自动评估的可行性研究
- Author:
Shouyi WEI
1
;
Xinying LI
;
Wei ZHANG
;
Shuo QUAN
;
Rongchao LIU
;
Xiaodong ZHANG
;
Jianxin LIU
Author Information
1. 北京大学第一医院医学影像科，北京 100034
- Keywords:
Radiation dose;
CTDI vol;
Segmentation;
Regression;
Deep learning
- From:
Chinese Journal of Radiological Medicine and Protection
2024;44(8):699-703
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To explore the feasibility of automatic assessment of abdominal and pelvic CT radiation dose index (CTDI vol) based on deep learning models. Methods:A retrospective analysis was conducted on clinical abdominal and pelvic CT data collected continuously from February 2021 to February 2022. A total of 1 084 sets of patient images were obtained using equipment of Siemens SOMATOM Definition Flash CT, Philips iCT, and GE lightspeed VCT. The volume CT dose index (CTDI vol) prediction model consisted of two functional modules: organ segmentation and dose prediction. Based on the result of actual scanning area segmentation in the abdominal and pelvic area, CTDI vol was evaluated automatically by dose regression prediction module. The images of 1 084 patients included in the study were randomly divided into a training set of 784, a validation set of 196 and a test set of 104. Dice coefficient was used to evaluate the abdominal and pelvic segmentation performance of the hybrid model, and accurate number proportion and root-mean-square logarithm error (RMSLE) were used as the evaluation index of the CTDI vol estimation model performance. Results:In the test set, the Dice coefficient of the deep learning model in the task of CT abdominal image segmentation was as high as 0.998, and the RMSLE of the CTDI vol regression model in estimation of radiation dose was 9.41%, with an accuracy rate of 92%. Scatter plot analysis showed that some CTDI vol estimates had significant errors, indicating that the model might need to be further optimized in these situations. Conclusions:The deep learning models can accurately and automatically segment CT abdominal images and estimate radiation dose, which can be used for clinical radiation dose monitoring and management.
