ESTRO 2024 - Abstract Book

S4650

Physics - Optimisation, algorithms and applications for ion beam treatment planning

ESTR0 2024

ΔNTCP for Acute Cardiac Events (ACE) is 2.0% or more [1]. To this purpose, both a photon and proton plan are created, from which the NTCP gain is calculated. This workflow adds extra time to the patient preparation phase and additional workload to already limited treatment planning staff. The added value of automation has been recognized, and multiple studies have been performed on automated treatment planning for proton therapy ([2]-[4]). However, to our knowledge so far, no results were published on automated IMPT treatment planning for proton therapy in clinical practice for breast cancer

This study’s purpose is to introduce an automated treatment planning workflow for breast plan comparison. We aim to reduce the planning time while preserving plan quality.

Material/Methods:

We have built a framework in our clinical treatment planning system environment (RayStation v10B, RaySearch, Stockholm) enabling treatment planning for plan comparison with minimal user interaction. We have automated the full workflow, including data import, data processing, plan setup, optimization, robustness evaluation and calculating the ∆NTCP between a proton and photon plan. Here we focus on the performance of the optimization for single sided breast with a single prescribed dose level of 40.05 Gy in 15 fractions. We use a strategy where the manual planning is closely mimicked, with 6 rounds of optimization in which the objectives on OARs are iteratively lowered. For 13 patients previously receiving a plan comparison, we compared the dosimetric results for the automatically generated versus the manually generated plans. We evaluated ∆NTCP based on the Dutch national proton therapy indication protocol [1]. The parameters evaluated are 1. ΔNTCP for ACE, 2. Mean Lung Dose (MLD) as an indicator for excess risk of secondary lung cancer and 3. Mean Breast Dose (MBD) as an indicator for excess risk for secondary breast cancer. In a separate analysis, the time required for plan generation for 20 patients was measured for both automated and manual planning, by recording the time between creation of help structures for treatment planning, and finalization of the final robustness analysis.

Results:

Plans were successfully automatically generated for the tested 13 patients. An example of the difference between automatically and manually generated treatment plan is shown in Figure 1. For 12 out of 13 patients, the conclusion on referral for proton therapy based on the NTCP comparison between photons and protons remained the same. In one patient, the ΔNTCP was 1.93% in the automatically generated plan, where the required ΔNTCP of 2.0% was exactly obtained in the manually generated plan. We do perform a manual review on all plans as part of our routine workflow, and based on this outlier we do recommend to look out for such cases in the review. The plans can subsequently be improved by manual continuation of the planning.

The automatically generated treatment plans were highly similar to the manually generated plans (Figure 2). Only the difference for MLD was statistically significant (p-value <0.05), in favor of the automated planning.

The time for generating the plans reduces by on average 1 hour, on an overall time of 3 hours required for the automated plan creation. During this time no manual action by the user is required for the automated plan creation, the manual planning does include idle time and iterative manual adjustment of the objectives.