Artificial Intelligence-Driven One-Click Automatic Treatment Planning for Functional Lung Avoidance Radiation Therapy: Clinical Implementation and Comprehensive Evaluation
Abstract
Purpose
To advance the clinical application of a fully automatic planning system for functional lung avoidance radiotherapy (AP-FLART) through clinical implementation and comprehensive evaluation.
Methods
AP-FLART integrates dosimetric score-based beam angle selection, multi-modality-guided dose prediction (MMDP), and function-guided dose mimicking. The system was implemented within the clinical treatment planning system RayStation. The MMDP model was pre-trained with 1,971 conventional lung radiotherapy (ConvRT) and FLART plans and then fine-tuned with 50 pairs of high-quality manual ConvRT/FLART plans. Clinical performance and potential benefits were assessed using a test dataset comprising 33 lung cancer patients who underwent SPECT ventilation or perfusion imaging and lung radiotherapy. Dosimetric metrics and normal tissue complication probabilities of automatic FLART plans generated by AP-FLART were compared with those of manual ConvRT and FLART plans created by an experienced planner. Additionally, three senior clinicians conducted blind reviews and comparisons of the automatic and manual FLART plans.
Results
Compared to manual ConvRT plans, automatic FLART plans significantly reduced the functionally weighted mean lung dose and high-function lung mean dose by 11.8% and 15.1%, respectively. Among FLART-benefiting patients, automatic FLART plans reduced the probability of grade ≥2 radiation pneumonitis by 6.25 percentage points (27%), while maintaining comparable probabilities for other side effects (all increases within 0.25 percentage points). Blinded reviews indicated that 87.9% of automatic FLART plans were clinically acceptable without modification, and 68.7% were rated as comparable (38.4%) or superior (30.3%) to manual FLART plans. Furthermore, AP-FLART reduced planning time from 2–3 hours for manual FLART planning to approximately 8 minutes.
Conclusion
A clinically viable, one-click auto-planning system for FLART (AP-FLART) has been successfully implemented and comprehensively evaluated. AP-FLART demonstrates substantial potential to enhance the efficiency, consistency, and quality of FLART planning, while reducing radiation-induced lung toxicity compared to the current clinical standard of lung radiotherapy.