Novel Method for More Efficient Optimizing the Knowledge-Based Planning: Specific Voxels of each Structure Influenced by Dominant Beamlets (SVSIDB)
There is a huge problem and time-consuming computation to optimize the IMRT treatment plan. Extracting the optimized plan from the predicted 3D3 so-called optimizing the KBP is also involved in this challenge. Some algorithms and methods have been presented for clustering and down-sampling the voxels to make the problem smaller, in recent years. In the current research, a novel down-sampling method is presented for optimizing the knowledge-based planning more efficiently. The concept of SVSIDB and corresponding down-sampling algorithm are proposed under title of SMP-2. The algorithm has been run on the data of 30 patients from the Open-KBP dataset. For each patient, there are 19 sets of dose prediction data in this dataset. Therefore, a total of 570 KBP-optimizing problems have been solved by applying the QuadLin model in the CVX framework. Resulted plans are evaluated and compared regarding two main fields which are the quality of the treatment plan as well as the computation efficiency. Solve time is the evaluation criteria for the latter field i.e. computation efficiency. The results of the current study indicated a remarkable improvement in the computation efficiency. Accordingly, the proposed method, SMP-2, reduced the average solving time by 46% in comparison to the full-data QuadLin model. The results also show an up to 53% reduction in solve time along with up to 22% improvement in clinical criteria compared to the previous research. Evaluation of the research results indicated that the SVSIDB has not only reduced the solve time but also improved the quality of the treatment plans. This is a remarkable achievement of the proposed model compared to the previous research and confirmed the significant effectiveness of the SVSIDB method which has the potential of even more improvement of the computation efficiency.