Development of a semi-automated superimposing image verification system using a template matching algorithm in radiotherapy

To analyze shifts in the isocenter of images, we developed a semi-automated superimposing image-verification system that is capable of automatically quantifying shifts in the isocenter through image analysis with a personal computer (PC). The accuracy and usefulness of this software were examined through a comparison of nine portal images with a simulation image and by comparing nine portal images with a DRR image, using a human pelvic phantom. The difference between the known magnitude of shift and the magnitude of shift detected with this method was analyzed as detection error. When the portal images were compared with the simulation image, the 95% confidence interval (95% CI) of detection errors (mean+/-SD) was 0.57+/-0.36 mm (95% CI: 0.49-0.65 mm). When the portal images were compared with the DRR image, the respective figures were 0.68+/-0.38 mm (95% CI: 0.59-0.77 mm). No significant difference was noted between these two categories of comparison (N.S). The absolute detection error (mean+/-SD) in all directions was 0.34+/-0.34 mm for the comparison of portal images with the simulation image and 0.41+/-0.36 mm for the comparison of portal images with the DRR image. This system seems to be appropriate for verification of the treatment field by improving the accuracy of radiotherapy as a method of computer-assisted landmark recognition during image comparison.