(Sunday, 7/29/2018) 3:00 PM - 6:00 PM
Room: Exhibit Hall
Purpose: Head and Neck (HN) cancer patients will greatly benefit from adaptive radiation therapy (ART) that takes into account patents’ weight lost and posture changes. However, labor intensive work and difficult decisions have limited the full-scale implementation of ART. This study proposes an adaptive planning information system (APIS) that solves this issue.
Methods: The APIS database is a central hub for storing/synchronizing demographic, dosimetric, imaging and response data for adaptive patients. Information is consolidated from three sources: Varian Eclipse TPS (initial and adaptive plans), and two custom applications for Longitudinal Imaging/Response Assessment (CVART) and Adaptive Decision Support. Patient demographic, prescription and planning data (including DVH statistics) are imported from Eclipse. Adaptive patients undergo initial and weekly MRI which is analyzed by CVART. Deformable registration between baseline and on-treatment MRI tracks tissue changes at each session with minimal user interaction and the observed volume changes are automatically imported into APIS. The Adaptive Decision Support application fetches geometric features from APIS continuously, uses machine learning (ML) to predict the cumulative dose and help decide whether to adapt. In this initial implementation, users must initially set up the patient in APIS and CVART, review the accuracy of the segmentation weekly, and enter certain data fields manually. Further automation is being developed including the ability to automatically send alerts when changes exceed established adaptive triggers.
Results: APIS currently includes data for 100 HN patients with a focus on parotid gland changes. Data import automation has reduced data entry errors which were 3-5% with manual entry. Automation has led to an average 30-40 minutes time savings for staff per patient.
Conclusion: APIS has reduced data entry errors, improved efficiency, and facilitated adaptive planning. Future improvements will include additional automation, improvements to machine learning and deformable registration algorithms, and inclusion of additional normal tissues and targets.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
