Nonrigid Image Registration for Detecting Temporal Changes on Thoracic MDCT Images
Image warping is a widely used technique for image registration that deals with geometric transformation techniques in the computer vision and image processing fields. It was introduced for geometric correction applications based on affine transformation, elastic deformation, and optical flow in remote sensing in the mid- 1960s. There are many techniques to detect opacities such as solid and/or GGO on thoracic computed tomography (CT) images .
The temporal subtraction technique  is used in medical imaging to emphasize subtle differences by subtracting a previous image from a later one. This technique can enhance interval changes such as new lesions and/or worsening existing abnormalities by subtracting the two image sets. Some commercial versions have been introduced.
In temporal subtraction, the image warping technique is the most important method for accurately deforming previous images to match current ones. When a misregistration occurs, image artifacts will appear on the subtraction images, often consisting of remaining normal structures such as blood vessels or airways. Ishida et al.  proposed a 2D image warping method in chest radiographs for CAD. However, in the temporal subtraction with MDCT, it is necessary to employ a more complex and accurate 3D registration scheme. To overcome this problem, we propose a new technique for automatic image warping to reduce subtraction artifacts on temporal subtraction . In this section, we describe the new temporal subtraction techniques and its application to detect abnormalities such as lung nodules on MDCT images. In the first step for image registration, we use a global image matching technique to correct for the global displacement caused by variations in patient positioning by use of a 3D cross correlation technique. In the second step, a local matching technique and a 3D elastic matching technique are used on the volume of interest (VOI). In the third step, a voxel matching technique is applied to register the two image sets. Finally, temporal subtraction images are generated by subtraction of the previous image from the current image. We have applied our computerized scheme to 31 MDCT image sets, which include examinations performed at two time points.
K. Mori et al.