(Wednesday, 7/17/2019) 9:30 AM - 10:00 AM
Room: Exhibit Hall | Forum 5
Purpose: Large GTV-nodal masses commonly show dramatic shrinkage during radiation treatment. When the nodal GTV is adjacent to organs at risk (OARs) such as parotid, dose delivered to OARs may deviate from the planned dose. We have developed an automated method to track and detect significant nodal volume shrinkage during treatment.
Methods: Pre-treatment and weekly T2 fat-suppression MRIs were acquired during the treatment course. GTV nodal contours were propagated from pre-treatment MRI to weekly MRIs using an automated deformable registration workflow using MIM™ software. Deformable registration accuracy was limited due to significant nodal volume shrinkage towards the middle/end of the patients’ treatment course. As the result propagated nodal volume is no longer reliable for volume calculations. Instead, all MRIs were processed to obtain image saliency maps based on intensities and local orientations in a multi-resolution pyramid using a classical visual attention method to evaluate whether the nodal location has strong visual attention. The ratio of mean saliency value from a propagated weekly nodal contour to the mean saliency value of the pre-treatment contour was calculated to assess whether the nodal volume shrunk significantly (< 1 c.c.). Specificity and sensitivity tests were done using weekly nodal volumes drawn by a physician or a senior planner. Optimal saliency ratio threshold is identified by analyzing the AUC from the evaluation of 28 MRIs in 8 patient cases.
Results: The optimal saliency ratio was identified as 0.6 to achieve Sensitivity/Specificity of 0.956 and 1.0 respectively for detecting significant volume shrinkage.
Conclusion: Image saliency shows the potential of being a computationally inexpensive method for detecting significant nodal volume shrinkage. We will further investigate this method with more patient cases and the development of a trigger to alert MDs when a volume change is significant enough to warrant the need for adaptation.
Keywords
Image Analysis, Image Processing
Taxonomy
IM/TH- Image Analysis (Single modality or Multi-modality): Computer/machine vision
Contact Email
