Deformable image registration (DIR)
Provide high-precision non-linear 3D registration.
Achieve high-speed calculation!
Due to the complexity and high cost of DIR calculations, the problem was the increased computation time. To address this, parallel computing with CPUs, GPUs, and CPU/GPU hybrid computing was implemented to achieve faster speeds. Additionally, the computation algorithm was optimized for further acceleration.
Innovative error estimation method
The error evaluation of DIR is most easily done through visual evaluation using difference images. While quantitative evaluations such as ICE (inverse consistency error) have been proposed, they assume that the error of the inverse transformation of image registration is infinitely small. However, this product developed its own error evaluation method to represent errors on the image, achieving both quantitative and visual evaluations.
Semi-automatic control
By reading in the ROI shape data of the reference breathing phase and inputting it into this product, automatic calculation of ROI shapes for other breathing phases can be achieved. Furthermore, with its excellent user interface, natural modification of ROI shapes can be performed.
(Left) Difference image of expiratory phase image and inspiratory phase image. (Right) A differential image of an image obtained by registering an image of the inspiratory phase to the expiratory phase by DIR and an image of the expiratory phase.
Black and white indicate differences between the two images. Gray indicates no difference between the two images. Respiratory movement reveals different positions/shapes.
By displaying DIR errors in a color map, DIR errors in areas related to treatment can be checked instantly.
The upper row shows the expiratory phase CT images, and the lower row shows the inspiratory phase CT images.vinegar. We calculated the deformation amount (deformed vector field: DVF) between both image data, and calculated the target ROI (yellow line) for the expiratory phase and the target ROI for the inspiratory phase.. You can see that the target shape is deformed.
In addition, by inputting the dose distribution for each respiratory phase, it is possible to calculate the time-accumulated dose distribution by combining the distribution with the reference phase.
The amount of movement for each respiratory phase can also be evaluated for each respiratory phase.
When calculating the dose distribution for each respiratory phase using 4D CT images (left), regions that are irradiated and regions that are not irradiated are produced depending on the time. In addition, regions where the dose varies with time may also occur even if they were irradiated. With only the distribution at each time point, a comprehensive dose evaluation cannot be performed. It is necessary to determine how much dose is irradiated for each voxel over the entire time. However, tracking the position of the same voxel is challenging because the shape and position change with time, as can be seen by watching the video. Therefore, the amount of deformation of the CT images at each time is calculated by DIR, and the dose distribution is transformed to the reference phase (middle). From here, all that is left is to calculate the time-integrated dose distribution (right) by weighting along the time axis (in the case of heavy particle radiation, other considerations may also be necessary). The accuracy of DIR is crucial for this dose distribution, and high-precision DIR is required.