Estimation of Cumulated Activity Images by Simultaneous Reconstruction of SPECT Projection Data from Multiple Time Points

 

Na Song and Eric C. Frey

 

PURPOSE:

In treatment planning for targeted radionuclide therapy (TRT) for cancer it is desirable to give patients a radiation dose that is high enough to kill tumors but not high enough to cause adverse effects in normal organs. However, to date, there has only been weak evidence of a dose-response TRT, making predictions of response or adverse effects difficult. One possible explanation is that non-uniform distribution of the therapeutic agent inside tumors or normal organs, resulting in non-uniform dose-distributions. These non-uniform dose-distributions are not handled well by standard phantom dosimetry methods such as the MIRD schema. Voxel-based dosimetry methods have been developed that would allow estimation of dose-volume histograms and could provide a better predictions of a dose-response relationships. These methods require as an input a 3D map of the cumulated activity, i.e., the integral of the time-activity curve (TAC). The purpose of this work is to develop reconstruction methods that allow estimation of this 3D cumulated activity distribution based on SPECT imaging performed at multiple time points.

 

METHODS AND MATERIALS:

There are several difficulties in developing such a method. First, even with careful patient positioning, the position of the patient and the internal organs and tumors will be different during the different SPECT imaging sessions. Second, noise in the SPECT images will result in non-uniformities in the reconstructed images and make accurate estimation of the cumulated activity in voxels difficult or imprecise. To overcome this we are developing a reconstruction code that takes into account the misregistration and directly obtains an image of cumulated activity distribution. The first step will be to register the images. While CT images are available and provide higher resolution than the SPECT images, they are non-contrast images so the contrast between important organs and the background is less than ideal. Further, the CT images, which are obtained on SPECT/CT system based on the GE Hawkeye CT, are obtained using slow rotation and thus contain artifacts from breathing motion during the scan. We thus propose to jointly register the CT and SPECT images obtained at the various time points to each other. We are investigating mutual information between both the CT and SPECT images as a measure of the degree of registration. Initially we will investigate rigid rotation of individual organs, but it may be necessary to include non-rigid rotation. To reduce the impact of image noise on the cumulated activity images, we propose to combine the reconstruction of the images with the fitting of the TACs. This will be done by maximum a posteriori (MAP) reconstruction of the images using constraints based on the fitted TACs with fitting of the TACs in each voxel from reconstructed images.

 

RESULTT

We began to investigate mutual information-based image registration methods, which begins with the estimation of the joint probability of the intensities of corresponding voxels in the two images. These methods differ in terms of the comparative ease of matching intensity images across modalities and the ability to discard intensity patterns in either modality that are not relevant to registration. We are considering the implementation of these methods ourselves or the use of existing class libraries such as the image Insight Segmentation and Registration Toolkit (ITK).

 

CONCLUSIONS:

We believe that the combination of registration of SPECT studies from multiple time points combined with direct reconstruction of the cumulated activities in voxels in the images has the potential to provide the necessary data to perform improved dosimetry using voxel-based dose estimation methods. Once developed, these methods will be applied to data from clinical dose-response studies to search for a robust dose-response.

 

FUNDING SOURCES:

This work is funded by Public Health Service Grant R01-CA109234.