Implementation of Machine Learning-Based Treatment Planning Tool for Whole Breast Radiotherapy Using Irregular Surface Compensator Technique

Abstract

To implement the machine learning (ML)-based treatment planning tool and compare plan quality and planning efficiency with the manual treatment process for whole breast radiotherapy using irregular surface compensator technique. Thirty whole breast or chest wall cases were planned by using the irregular surface compensator technique with fluence maps manually and iteratively edited to achieve uniform dose distribution to the target. Patients were treated with these manual plans after physician’s approval. For each case, an in-house ML-based automated treatment planning tool was implemented in a programming interface and generated the fluence maps with the same beam parameters such as beam energy, gantry angle, collimator angle, and aperture shape as manual plans.An analytical algorithm software with 0.25cm grid size was used in dose calculation for all plans. Breast or chest wall clinical target volume (CTV) coverage based on the percentage CTV volume receiving 95% of the prescribed dose (V95%) and high-dose volume based on V105% were compared to evaluate the plan quality as well as the planning efficiency. Two-tailed Wilcoxon Signed-Rank test was performed to test the null hypothesis that the two planning schemes yield equivalent plan quality. The mean planning time was 110.2 min with standard deviation (SD) of 62.8 min for the manual planning with the range from 25 to 270 min, and 6.4 min with SD of 2.1 min for the ML-based planning with the range from 4 to 12 min (p<0.01). CTV mean V95% was 96.7% (SD: 5.0%) for the manual planning and 96.7% (SD: 4.8%) for the ML-based planning (p=0.89). CTV mean V105% was 21.6% (SD: 29.8%) for the manual planning and 20.4% (SD: 30.5%) for the ML-based planning (p=0.22). Differences in doses to heart and lungs were negligible between the paired plans as the two planning schemes used the same beam parameters. This study successfully implemented the ML-based automated treatment planning tool through scripting software. Abiding to the same plan quality as manual process, the automated tool significantly reduced the planning time as the ML-based tool automate the iterative fluence editing process. This technique demonstrated promising potential to significantly improve efficiency of the treatment planning for whole breast or chest wall radiotherapy.