A novel mathematical model to generate semi-automated optimal IMRT treatment plan based on predicted 3D dose distribution and prescribed dose.

Abstract

In recent years, with the development of artificial intelligence and deep learning techniques, it has become possible to predict the three-dimensional distribution dose (3D3 ) of a new patient based on the treatment plans of similar recent patients. Therefore, some new questions have arisen for the above issue: how to make use of the predicted 3D3 obtained from deep learning, to facilitate treatment planning? How to convert the predicted 3D3 to a clinical deliverable Pareto optimal plan? Little research has been done and limited software has been developed in this regard. In the current research, an attempt was made to contribute the knowledge-based planning by presenting a new mathematical model, and to take a novel step towards optimizing the treatment plan derived from both predicted 3D3 as well as dose prescription to generate a semi-automated clinically applicable optimal IMRT treatment plan. The presented model has benefited from both prescribed dose as well as predicted dose and its objective function includes both quadratic and linear phrases, so it was called the QuadLin model. The model has been run on the data of 30 patients with head and neck cancer randomly selected from the Open-KBP dataset. There are 19 sets of dose prediction data for each patient in this database. Therefore, a total of 570 problems have been solved in the CVX framework with commercial solver Mosek and the results have been evaluated by two plan quality approaches (1) DVH points differences, and (2) satisfied clinical criteria. The results of the current study indicate a strong significant improvement in almost all plan evaluation indicators compared to the reference plan of the dataset, 3D3 predictions, as well as the results of previous research, based on the Wilcoxon signed ranks test with a significance level of 0.01. Accordingly, for all regions of interest (ROIs) (or structures) of all 570 problems total clinical indicators have improved by more than 21%, 15%, and at least13%, on average, compared to the predicted dose, the reference plan, and previous research, respectively, with 341 s as the average of solving time. Evaluation of the research results indicates the significant effect of the QuadLin model on improving the dose delivery to the target volumes while reducing the dose and preserving organs at risk. Based on the literature, the proposed model has generated the best-known treatment plan from the predicted 3D3 so far.