Radiation Physics

PV QA 3 - Poster Viewing Q&A 3

TU_8_3187 - Application of Machine Learning Techniques for Accurate Dose Calculation of Electron Treatment with Small and Irregular Fields

Tuesday, October 23
1:00 PM - 2:30 PM
Location: Innovation Hub, Exhibit Hall 3

Application of Machine Learning Techniques for Accurate Dose Calculation of Electron Treatment with Small and Irregular Fields
X. Zhou1, L. Xing2, and B. Han2; 1Tsinghua University, Beijing, China, 2Stanford University School of Medicine, Palo Alto, CA

Purpose/Objective(s): Electron treatment with Small and Irregular Fields widely used in radiation therapy. The current technique is to perform film or ion chamber measurement to confirm output factor(OF), which is time consuming and prone to errors. The purpose of this study to apply state-of-the-art machine learning techniques for accurate dose calculation for small and irregular field electron treatment.

Materials/Methods: The Linac electron beam commissioning and clinical measurement data were collected from our institution. The datasets were then randomly split into training and testing data. The measured dose output factors were treated as outcomes of linear regression estimator and 10-fold cross validation model with inputs of different cone size, 2D cutout map, and SSD. Using data augmentation technique, the number of training samples were increased by adding randomized Gaussian noise to the original datasets. Dose output factors were predicted using models trained with regularization added to the cost functions. The predicted dose output factors for small and irregular electron treatment fields were evaluated using percentage relative error regarding measured data at the depth of Dmax. A special treated R2 metric were used to evaluate to what extent the model explains the changes of variables. The best possible R2 score is 1.0.

Results: The package we used was scikit-learn in python. With augmentation technique, a total of 445 data samples were tested for model training and dose output prediction. The dose output factors for small and irregular electron treatment fields were accurately predict with the proposed machine learning methods. The average relative absolute error between the predicted and measured electron dose output factor is 1.57%. The R2 metric evaluation of the model is 0.994.

Conclusion: Our results showed that our proposed machine learning method produced more accurate dose distribution as compared conventional dose calculation methods. The improved dose accuracy of the propose approach will significantly shorten the conventional tedious output measurement and has the potential to enhance the treatment efficacy.

Author Disclosure: X. Zhou: None. L. Xing: Research Grant; Varian Medical Systems. Honoraria; Varian Medical Systems. Royalty; Varian Medical Systems, Standard Imaging Inc. B. Han: None.

Bin Han, PhD

Disclosure:
Employment
Stanford University: Clinical Assistant Professor: Employee

Presentation(s):

Send Email for Bin Han


Assets

TU_8_3187 - Application of Machine Learning Techniques for Accurate Dose Calculation of Electron Treatment with Small and Irregular Fields



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Application of Machine Learning Techniques for Accurate Dose Calculation of Electron Treatment with Small and Irregular Fields