MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Developing and Validating An Automatic System of Tracking Parotid Shrinkage in Weekly MRIs for Adaptive Radiotherapy

Y Hu*, S Fontenla , N Tyagi , J Mechalakos , P Zhang , C Polvorosa , N Allgood , G Mageras , M Hunt , Memorial Sloan-Kettering Cancer Center, New York, NY

Presentations

(Tuesday, 7/31/2018) 10:30 AM - 11:00 AM

Room: Exhibit Hall | Forum 2

Purpose: Adaptive radiotherapy(ART) is generally a resource intensive process. Tools to select patients who may benefit most from plan adaptation during treatment become increasingly valuable. We have developed a system that automatically monitors parotid shrinkage and alerts clinical team during the treatment for head-and-neck patients. We report the validation results in this study

Methods: We acquired a T2 fat-suppression MRI in the patient’s treatment position and in the more comfortable diagnostic position on the simulation day as the baseline and weekly T2 fat-suppression MRIs in diagnostic position to monitor anatomical changes. A nightly process in the system queries and retrieves data via DICOM, registers planning CT to baseline MRIs, and when a weekly MRI available, registers the diagnostic baseline MRI to the weekly MRI. We used a mutual information based deformable registration method in Plastimatch (MGH) for CT-MR registration and a customized deformable registration workflow in MIM™ for MR-MR registration. Parotid contours were propagated to weekly MRIs using the registrations and pushed back to MIM via DICOM by the nightly process for validation by treatment planning personnel.To compare the parotid contours automatically generated by the system with the user-validated contours, we calculated the percentage difference in parotid volumes, DICE similarity coefficient and 95% Hausdorff distance. We evaluated 133 baseline and weekly MRI.

Results: The mean(±s.d.) of the parotid volume difference between auto-generated and validated contours is 9.02%(±6.9%) and the median is 9.01%. The mean of the DICE similarity coefficient is 0.87(±0.10) and the median is 0.88. The mean of the 95% Hausdorff distance is 4.50mm(±3.1mm) and the median is 4.00mm.

Conclusion: Our fully automatic system accomplished the resource intensive registration tasks without the need of human intervention and demonstrated the capability to accurately monitor the parotid shrinkage in H&N treatment.

Keywords

Computer Software, Image Analysis, Registration

Taxonomy

IM/TH- Image Analysis (Single modality or Multi-modality): Computer-aided decision support systems (detection, diagnosis, risk prediction, staging, treatment response assessment/monitoring, prognosis prediction)

Contact Email