Rule-based AI automated adaptive treatment planning for image guided cervical cancer brachytherapy
Abstract
BACKGROUND AND PURPOSE
A rule-based AI system for automated adaptive treatment planning for image guided adaptive brachytherapy (IGABT) of locally advanced cervical cancer (LACC) was developed at Erasmus MC, and internally and externally validated by Erasmus MC and Tata Memorial Centre (TMC), respectively.
MATERIALS AND METHODS
The BiCycle system generates automated plans with adapted requirements for each fraction, considering previously delivered external beam radiotherapy (EBRT) and BT doses, according EMBRACE-II protocol. It optimizes dosimetric parameters and loading patterns for available radioactive source positions. The system’s effectiveness was validated by comparing automatically generated plans (AUTO) with manually generated, clinically delivered plans (MANUAL) of (1) dosimetry parameters and loading pattern visual inspection of 15 previously treated patients, for internal validation and (2) dosimetry and qualitative comparison by two TMC physicians of 20 previously treated patients, for external validation.
RESULTS
With comparable target doses, AUTO plans had reduced D2cm3 (expressed as EBRT + BT total EQD2α/β) for bladder, rectum, sigmoid and bowel compared to MANUAL plans with average gains of 5.3 Gy, 2.4 Gy, 2.5 Gy and 2.7 Gy, respectively, for internal validation, and of 3.6 Gy, 4.3 Gy, 1.9 Gy and 1.1 Gy, respectively, for external validation. The two TMC physicians preferred the AUTO plans in 76.3% and 75.0% of comparisons.
CONCLUSION
A novel AI-system for fully automated IGABT treatment planning for LACC allowed high-quality plan optimization in on average 1.6 min. AUTO plans were considered superior in quality compared to MANUAL plans in both internal and external validations, even without optimizing the system’s configuration for the external center.
Keywords
- Automation
- AI
- Image guided adaptive brachytherapy
- IGABT
- Automated treatment planning
- Locally advanced cervical cancer
- LACC