123I-FP-CIT is used for the diagnosis of Parkinson's disease (PD) and Dementia with Lewy Bodies (DLB) [1]. It can visualize the distribution by binding to the dopamine neurons. In negative cases, the dopamine neurons are accumulated in form of comma, and the dopamine neurons are accumulated in form of dot in positive cases [2]. SPECT images can provide the functional information of a brain, but it is not possible to show enough anatomical information such as shape of corpus striatum. Therefore, presenting brain MR images on the SPECT images may improve the diagnosis accuracy and quantitative evaluation based on the uptake values on SPECT images. The image fusion and registration is a common technique for the diagnosis [3-6], since multi modality images are used for the diagnosis of dementia. The purpose of this study was to develop the fundamental image registration technique for quantitative image analysis of brain SPECT images to obtain precise uptake values on SPECT images by using object segmentation results on MR images.

Since image information represented by SPECT and MR images are different each other, an accurate image fusion of the two images is not easy. In this study, we propose a new image fusion method for SPECT and MR image by using very low dose CT images for absorption correction in SPECT/CT devices. The CT images were utilized for the image registration with MR images.

Figure 1 shows the overview of our processing steps. The uptake values on SPECT images were measured based on the segmented regions determined on MR images after the image registration were performed.

Figure 2 shows the registration process in this study. The location of CT and SPECT images were mechanically registered because the images were obtained at the almost same time. For the image registration between CT and MR images, we used a global matching by using center point and volume s after the image registrations were matched. The mutual information (MI) between the two images was used for the fine registrations [4-6].

The MI is the amount that represents a measure of the mutual dependence of two variables. We unified the spatial resolutions of CT and MR images changed from 0.4297 mm to 2.9454 mm.The maximum MI was obtained with various locations and angles to determine the final parameters of the location and the angles. Parallel computing method using CUDA was also applied to obtain the parameters.After the image fusion between the CT and the same patient's MR images was completed, the region locations and the area on MR images could be mapped on the SPECT images.

We employed 30 cases of SPECT/CT and MR cases for the evaluation. 25 of 30 cases were registered correctly with registration errors less than 5mm. These results enable to measure precise uptake on SPECT images based on the segmentation results on MR images. The regions manually set on MR images could be mapped to obtain the uptake values on the SPECT images.

Figure 3 shows an example for the registration result. The MR/SPECT image was obtained by the registration result of CT/MR. The high uptake values on the SPECT image were mapped on the MR image correctly as dot shapes on corpus striatum. Figure 4 shows a histogram of registration errors. The minimum and maximum errors were 1.2 mm and 16.2mm, respectively. In most cases, the error was less than 5 mm. Figure 5 shows an overview of our developed software by using three modality images.

The registration errors in these clinical cases were larger than those in a phantom study to estimate the mechanical errors. The registration errors in the phantom study were less than 3 pixels (1.3 mm). The errors of approximately 5mm in clinical cases were larger than the phantom results, but the errors could be ignored because the pixel size of SPECT image is 3 to 5 mm per pixel.

We proposed a new approach to register brain SPECT and MR images by calculating mutual information of low dose CT images for absorption correction taken by SPECT/CT and MR images. The registrations were performed with small errors less than 5mm. The shape information obtained by MR images could be superimposed on SPECT images based on our registration method.