Quantitative Image-Based Registration Metrics for Re-Irradiation Validation
Abstract
Purpose
Dose summation for re-irradiation requires a transformation of the dose from the previous plan on CT data set CT1 onto a new plan on CT data set CT2. The dose transformation corresponds to the CT1-CT2 image transformation. Therefore, the validity of the sum dose depends directly on the validity of the image transformation. Many authors have noted that quantitative image registration metrics essential to validate the corresponding plan sum are often lacking in commercially available products performing deformable image registration. This work presents a platform to compute four metrics without user input on a 0 – 100 scale suitable for physician review.
Methods
CT1 and CT2 dicom data sets were imported into MATLAB with the imaging toolbox. The images can be cropped to analyze a specific region of interest. A representative bladder case study which qualitatively appeared unsatisfactory was chosen to demonstrate the process using standard Structural Similarity (SSIM), Cross-Correlation (CC), Mean Square Error (MSE), and an innovative Hounsfield Unit Difference (HUD) metric indicating the mean percentage of points with a CT1-CT2 difference less than 20%. The metrics were presented as a Trustworthiness (T) bar graph including the results for both Rigid and Deformable registration as well as their difference as an analytical measure of deformation.
Results
SSIM (T=68.4%) which includes luminance, contrast, and structure is the recommended comprehensive clinical metric. Absolute HU difference (T=41.9%) is very sensitive and but it is best used in conjunction with a positional display. A vivid bar graph presentation of all metrics is useful for physician review.
Conclusion
Deformable image registration is prone to non-unique solutions and therefore plan sums should be presented with an indication of trustworthiness. MATLAB provides users with the ability to compute and present relevant vendor neutral quantitative metrics effectively so that physicians can perform an informed evaluation