Advertisement

A Rapid Learning Approach to the Knowledge Modeling of Radiation Therapy Planning

      Purpose/Objective(s)

      The purpose of this study was to implement a rapid learning method to train knowledge models to predict the organ-at-risk (OAR) dose sparing in radiation therapy (RT) based on an array of patient anatomical features. We also aimed to establish the evaluation criteria and validation method to ensure the accuracy and the efficiency of the learning process.

      Materials/Methods

      The knowledge models to predict OAR dose sparing have been shown to be useful tools to guide RT planning. A rapid learning approach was utilized to train the knowledge models in this study. Rapid learning mimics human knowledge accumulation. A base knowledge model was first established from an existing database. Then each subsequent case was used to continuously evaluate and update the knowledge model. A total of 100 clinical cancer cases in the pelvic region were retrospectively analyzed. Among them, 40 cases were low to intermediate-risk prostate cases (type I), 20 were high-risk prostate cases with lymph node irradiation (type II), and 40 were anorectal cancer cases (type III). Starting from a base model for type I cases, an increasing number of cases with more complex planning target volume (PTV)-OAR anatomies (type II and type III) were continuously added into the training case pool. The studentized residual and the leverage values were calculated as evaluation criteria at each step, which were used to identify outliers and to discriminate if a large prediction residue was due to plan quality variation or because the new case was an isolated case in the feature space. The efficiency and accuracy of the learning method was quantified by the learning curve, which was extracted by a boot-strap validation method. It described the longitudinal improvement of model accuracies with increasing number of training cases. The model predicted generalized Equivalent Uniform Dose (gEUD) in the bladder and rectum were compared with the actual values in the validation cases. The Median Absolute Differences (MAD) between the predictions and the clinical values were calculated.

      Results

      The MAD of the predicted OAR gEUD in all three types of cases gradually decreased when increasing number of training cases were added in training. The MAD of the bladder and rectum gEUD in both type II and III validation cases reached a stable value of 2.1% to 3.5% of prescription dose when 12 type II or type III were added in training, in addition to the 30 type I cases, and they were comparable with the MAD value of 2.0% to 3.4% when all cases were used to train the models in a batch mode.

      Conclusion

      The rapid learning approach is able to learn knowledge models for multiple cancer types in the pelvic region with comparable accuracy to the batch training method and with improved efficiency. This approach will facilitate the implementation of knowledge-based radiation therapy planning in clinics.

      Comments

      Commenting Guidelines

      To submit a comment for a journal article, please use the space above and note the following:

      • We will review submitted comments as soon as possible, striving for within two business days.
      • This forum is intended for constructive dialogue. Comments that are commercial or promotional in nature, pertain to specific medical cases, are not relevant to the article for which they have been submitted, or are otherwise inappropriate will not be posted.
      • We require that commenters identify themselves with names and affiliations.
      • Comments must be in compliance with our Terms & Conditions.
      • Comments are not peer-reviewed.