Clinical implementation and evaluation of deep learning-assisted automatic radiotherapy treatment planning for lung cancer

Abstract

Abstract Background and purpose The purpose of this study is to investigate the clinical application and assessment of deep learning (DL)-assisted automatic radiotherapy planning for lung cancer patients.Materials and methods We developed a deep learning model for predicting patient-specific dose that was trained and validated on a dataset of 235 lung cancer patients, and the model was integrated into clinical workflow to assist planners in generating treatment plans. We retrospectively selected and recovered additional 50 clinically treated lung cancer patients’ manual volumetric modulated arc therapy (VMAT) plans with different target volumes and different treatment patterns. Subsequently, automatic plans were generated for each of these patients. Both automatic and manual plans were subsequently compared in terms of overall plan quality metric (PQM), target coverage and homogeneity, organ at risk (OAR) sparing, monitor units (MUs), and planning time. Additionally, qualitative reviews of automatic and manual plans were implemented by four expert reviewers to assess the clinical applicability of DL-assisted automatic plans.Results The average PQM score was 40.7 ± 13.1 for manual plans and 40.8 ± 13.5 for automatic plans, and they had equivalent overall plan quality. The targets coverage and homogeneity of the automatic plans were considered equivalent or superior when compared to manual plans. Both plans had their own advantages in OAR sparing, such as better sparing of lung for manual plans and better sparing of heart for automatic plans. It is worth to note that the average planning time of automatic plans was reduced from 103.1 ± 18.5 minutes to 32.6 ± 5.3 minutes (P<0.001) and the MUs were reduced from 789.9 ± 234.3 to 692.5 ± 210.7 (P<0.001). In qualitative evaluation, automatic plans were deemed to be clinically acceptable for treatment in 88% of reviews (176/200), and all were accepted after fine tuning. Most expert reviews indicated a preference for equivalence between automatic and manual plans when making their selection.Conclusion The DL-assisted lung cancer plans demonstrated comparable or superior quality to manual plans, improved planning and treatment efficiency, and significantly reduced planning time and MUs. It has the potential to enhance the workflow of radiotherapy departments, ultimately providing tangible benefit to lung cancer patients.