Program Information
Database Plan Search to Facilitate Automated Planning of SRS Brain Metastases
E. Schreibmann*, E. Elder , Department of Radiation Oncology and Winship Cancer Institute of Emory University
Presentations
TH-CD-205-2 (Thursday, August 3, 2017) 10:00 AM - 12:00 PM Room: 205
Purpose: Knowledge-based planning has rapidly gained clinical acceptance but to date the approach is applicable only to optimization constraints. We propose a broader approach established by searching previous treatments for cases of similar anatomy and applying all planned parameters to the new patient.
Methods: The approach was implemented to aid treatment planning for brain SRS cases. The key to the proposed approach is to quantify the target shape and location relative to critical structures through a compact mathematical formulation allowing the comparison of hundreds of cases in a few seconds by a condensed representation of the shape characteristics. To construct the search engine, the segmentation and plans of 248 patient treated in our institution were saved to a dedicated server. All cases were rigidly registered to a common reference by alignment of the brainstem, optic nerves, and globes. Once aligned, the PTVs were resampled to the same number of points. Variance for each point was quantified to obtain a shape’s eigenvectors describing its discrepancies from the reference dataset. When planning a new case, the procedure is repeated and it’s corresponding eigenvalues are compared to each database case using a Mahalanobis distance quantifying their similarities.
Results: The search algorithm was integrated with the treatment planning system using scripting. The user can view the details of the search, such as the beam angles, DVHs and optimization constraints used previously on similar cases. For convenience, this information can also be formatted as a new DICOM plan, importable into the system through standard methods.
Conclusion: A search engine was created and customized to the specifics of SRS brain planning to retrieve planning aspects from cases of similar anatomy. The algorithm takes less than 1 minute to search and is integrated in the clinical software for streamline use in day-to-day clinical practice.
Contact Email: