MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Evaluation of Machine Learning Algorithms for Treatment Planning Parameter Calculation

J Chow1*, R Jiang2, F Ng3, (1) Princess Margaret Cancer Centre, Toronto, ON, CA, (2) Grand River Hospital, Kitchener, ON, CA, (3) Ryerson University, Toronto, ON, CA

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: Different machine learning algorithms were used to calculate the dose-volume parameter used in external beam treatment planning quality assurance. The performances of all algorithms were compared in term of their accuracy and calculation efficiency.


Methods: Dose distribution index (DDI), containing information of dose coverage conformity and homogeneity for the planning target volume, organs-at-risk and remaining target-at-risk, were calculated using machine learning. Machine training was carried out using dose-volume histograms of fifty prostate volumetric modulated arc therapy plans and algorithms, namely, linear regression, tree regression, support vector machine (SVM) and Gaussian process regression (GPR). The original calculation of DDI was also carried out, involving mathematical formulas and all dose-volume histograms of the targets and critical organs. To compare the performance among the machine learning algorithms, root mean square error (RMSE) showing the deviation between the DDI result from formula and machine learning, prediction speed and training time were determined.


Results: For the RMSE value, all algorithms regarding linear regression, SVM and GPR performed well with RMSE ranging from 0.0038 to 0.0193. The RMSE values of DDI using the medium and coarse tree regression algorithms, however, were found larger than 0.03. This showed that the tree regression algorithms were not suitable for calculating the DDI value. Comparing the RMSE, prediction speed and training time for all machine learning algorithms, it is found that the square exponential GPR algorithm had the smallest RMSE, relatively high prediction speed and short machine training time of 0.0038, 4,100 observation/s and 0.18 s, respectively.


Conclusion: It is concluded that machine learning can be used to calculate dose-volume parameter for treatment planning evaluation and quality assurance. Comparing the performance among different algorithms, it is found that the square exponential GPR is most suitable to predict the DDI value in this study.

Keywords

Dose Volume Histograms, Quality Assurance, Treatment Planning

Taxonomy

TH- Dataset analysis/biomathematics: Machine learning techniques

Contact Email