Room: Exhibit Hall | Forum 7
Purpose: To compare the robustness of three popularized dose-volume-histogram (DVH) prediction models for knowledge based radiotherapy treatment planning (KBTP).
Methods: Three popularized KBTP DVH prediction models were selected and realized to perform comparisons, naming proposed by Zhu et al. in 2011, Lindsey et al. in 2012 and Satomi et al. in 2015 respectively. Calculations were based on an identical archived IMRT treated plan cohort, one part of them is used for training, and the others for evaluation respectively. Prediction model accuracy was compared by calculating the mean sum of residuals (SR) between predictions and actual plan values. Robustness comparison was performed by observing the prediction accuracy changing as with training example number verse evaluation example number (TE#) ratio changes. A cohort of 50 retrospective selected prostate IMRT plans and 29 nasopharyngeal carcinoma (NPC) VMAT cases were used to perform this study respectively. DVHs of the bladder, rectum and parotid gland were inspected. TE# varied from 25:25, 30:20, 35:15, 40:10 to 45:5 on prostate cases and 23:6 for NPC cases.
Results: Results show Zhuâ€™s prediction model has slightly superiority over Satomiâ€™s and then over Lindseyâ€™s from the observation of mean SR, with corresponding values of 0.0677, 0.1663 and 0.0554 respectively for bladder, 0.0664, 0.1280 and 0.0855 respectively for rectum, 0.0268, 0.0267 and 0.0346 respectively for right parotid and 0.0282, 0.0298, 0.0368 respectively for left parotid. As the number of training cases increases, mean SR of Zhuâ€™s, Lindseyâ€™s and Satomiâ€™s method ranges [0.0577,0.0749], [0.1596,0.1931], [0.0540,0.0572] respectively for bladder, and [0.0605,0.0686], [0.1243,0.1335], [0.0803,0.0899] respectively for rectum, showing Zhuâ€™s and Satomiâ€™s prediction model have comparative performance but slightly prevail over Lindseyâ€™s by standing on a model robustness viewpoint.
Conclusion: The comprehensive results of the three methods in two cohorts illustrate that the Zhuâ€™s method had better performance and robustness than the other two methods.
Funding Support, Disclosures, and Conflict of Interest: 1) National Key R&D Program of China (NO.2017YFC0113203); 2) National Natural Science Foundation of China (NO.81571771 and 81601577); 3) Post-doctoral Science Foundation of China (NO.2016M592510). 4) Public Welfare Research and Capacity Building Special Foundation of Guangdong, China (2015B020214002)
Not Applicable / None Entered.