Case-Based Reasoning Optimization Algorithm for Inverse Planning in Head and Neck Radiotherapy
R Reiazi*, S Prajapati, A Mohamed, C Fuller, M Salehpour, UT MD Anderson Cancer Center, Houston, TX
Presentations
PO-GePV-T-260 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: The goal of a modern treatment planning system involves balancing competing clinical goals, a.k.a. optimization parameters when generating beam configuration. However, the process of tuning parameters is very time-consuming. It may compromise between tumor coverage and normal tissue toxicity such that, when a plan is accepted for treatment, it is not clear whether a better treatment plan exists. To mitigate these challenges, we introduce a Case-based reasoning approach to generate several radiotherapy treatment plans (RTPs), a.k.a Pareto-optimal plans, for a given patient based on optimization values from previous patients with an acceptable outcome.
Methods: Our proposed system has four independent components: 1-Preprocessing that includes a contour harmonization module to extract and rename contours from Dicom files, and a features extraction module to extract distance and shape descriptors from selected contours. 2-Retrieval to retrieve anatomically similar patients from our database using a hierarchical filtering algorithm. 3-Pareto-plan optimizer that generates and archives Pareto-plans. 4-Conversational agent allows users to decide which plan is clinically accepted by navigating through the archived Pareto-plans and optimization values. We use a patient cohort including ~5000 head and neck cancer patients to develop the proposed system.
Results: The mean dose conformity indices of the PTV and OAR contours with the highest weighting factor from selected patients out of ~500 laryngeal cancer patients were 1.68 (±0.3), 1.97 (±0.5), 2.12 (±0.9), and 2.3 (±1.1) for the first to the fourth similar patients compared to 1.15 (±0.07) in the original plans. All found matches had tumors on the same side as the given patients.
Conclusion: Our results show that the proposed system can generate acceptable RTPs from a set of proven, safe, and effective plans. By minimizing human interactions, this algorithm can generate RTPs in a fraction of the time that is currently required to achieve such results.
Keywords
Not Applicable / None Entered.
Taxonomy
TH- External Beam- Photons: IMRT/VMAT dose optimization algorithms
Contact Email