2262 - Deep Learning-Based Automatic Dose Optimization of Brachytherapy

02:30pm - 04:00pm PT

Hall F

Screen: 4

POSTER

Presenter(s)

Xianliang Wang, PhD - Sichuan Cancer Hospital & Research Institute, Sichuan, Sichuan Province

T. Liu^1,2, S. Wen^1,2, F. Zeng^1,2, and X. Wang³; ¹Applied Nuclear Technology in Geosciences Key Laboratory of Sichuan Province, Chengdu University of Technology, Chengdu, 610059, China, ²Sichuan Cancer Hospital & Institute, Sichuan Cancer Center, School of Medicine, University of Electronic Science and Technology of China, Radiation Oncology Key Laboratory of Sichuan Province, Chengdu, 610041, China, ³Sichuan Cancer Hospital and Research Institute, University of Electronic Science and Technology of China, Chengdu, China

Purpose/Objective(s):

This study aims to determine the optimal dose processing method in deep learning-based dose prediction for brachytherapy and to explore the feasibility of using the inverse dose optimization algorithm to further enhance the quality of treatment planning.

Materials/Methods:

Brachytherapy data of 186 patients with cervical cancer were collected retrospectively. The data were divided into a training set, validation set, and test set according to the ratio of 150:18:18. Square-root transformation normalization, logarithmic normalization, and linear normalization were applied to process the dose data, respectively. The predicted results were compared with the unprocessed dose data. The four groups of dose predictions were evaluated using the Dice similarity coefficient (DSC), conformity index (CI), and homogeneity index (HI). The group with the best overall performance was selected, and its dose prediction results were imported into a gradient-based planning optimization (GBPO) algorithm for further optimization. The target D90% was normalized to 6 Gy. The D1cc and D2cc of organs at risk were compared before and after optimization.

Results:

The dose prediction results with unprocessed doses had the best-combined performance on the DSC, CI, and HI metrics. The (DSC, CI, HI) values for unprocessed dose, square-root transformation normalized, log normalized, and linear normalized were (0.94, 0.74, 0.49), (0.93, 0.72, 0.50), (0.91, 0.71, 0.45) and (0.90, 0.71, 0.47), respectively. The predicted dose results of the unprocessed dose group were further optimized by the GBPO algorithm. The results showed that the (D1cc, D2cc) values for the bladder, rectum, and sigmoid decreased by (2.11%, 2.09%), (2.62%, 2.14%) and (3.16%, 2.98%), respectively, which were statistically significant (p<0.05). The dose of the small intestine increased slightly; the average increase in D1cc and D2cc doses was 2.08% and 1.63%, respectively, with no statistically significant difference (P>0.05).

Conclusion:

In deep learning-based dose prediction for BT, it is not recommended to preprocess the dose data. The predicted dose can be further optimized using inverse dose optimization algorithms to improve the quality of the treatment plan.