Keywords: Breast, Cancer Imaging.

Objectives: In a larger effort, we are studying methods to improve the specificity of the diagnosis of breast cancer by combining complementary information from multiple imaging modalities (PET, pre- and post-contrast MRI, Ultrasound, and Digital Mammography). Merging information is important for a number of reasons. For example, the determination of functional and corresponding anatomical locations in images from various modalities is necessary to determine the extent of disease. Furthermore, the time course of contrast uptake values are an indication of malignancy so the reduction of the influence of motion artifacts is important as well. To facilitate this fusion of multimodal information, image registration (alignment) becomes necessary. In this work, we describe our efforts to register 3D breast images from PET and MRI.

Methods: PET images are acquired with the patient in the supine position by a GSO-based Philips Allegro system (Philips Medical Systems, Andover, MA) and are reconstructed using 3D RAMLA with voxel sizes of 4x4x4 mm. Images sizes are typically 144x144x(45 to 170) slices (for breast or whole body scans, respectively). A 10 µCi/gram dose of FDG adjusted for patient weight is injected. Both transmission and emission images are acquired. Prone breast MRI images are acquired by a GE Signa 1.5T scanner (General Electric Medical Systems, Milwaukee, WI). T1 3D SPGR images are acquired using the standard GE software. Pre- and post-contrast (Gadolinium) images are reconstructed using a custom technique [1,2]. Voxel sizes for T1 images are 0.9x0.9x(2 to 4) mm with 256x256 images. The pre- and post-contrast images are 0.5x0.5x(2 to 4) mm and 512x384 pixels. Ten patients have been scanned from a population with suspicious lesions (ranging from 3x3x4 mm to 24x43x100 mm).

Results: Given these data sets (T1, pre- and post-contrast MRI, and transmission and emission PET) many comparisons can be made. For example, T1 MRI to transmission PET, pre-contrast MRI to emission PET, etc. Figure 1 illustrates the sequence of registration techniques as applied to the various modalities that we found to be most effective. Image registration techniques may be broadly classified into either rigid body (single, global transformation) or elastic/deformable (many local transformations) [3]. Two methods of rigid body registration (mutual information [4] and principle components analysis) are compared. A new semi-deformable registration technique (consisting of separate affine transformations for each breast) and the deformable method described in [5] are presented as well.

Conclusions: With varying degrees of accuracy, one may accomplish the difficult task of registering PET and MRI breast images. Our experiments indicate that Thirion's demons method performed best at registering MRI images acquired with various MRI protocols. For PET transmission and emission registration, Thirion's algorithm performed poorly and rigid body registration using correlation performed best. None of these methods performed well for MRI to PET transmission registration. Therefore, we developed a new method that we call pushpins that estimates two affine transformations (one for each breast) from only 6 points specified by the user in a T1 MRI and the corresponding PET transmission data set.

References
1. H.K. Song, L. Dougherty, K-space weighted image contrast (KWIC) for contrast manipulation in projection reconstruction MRI, Mag. Res. in Med. 44:825-832, 2000.
2. H.K. Song, L. Dougherty, M.D. Schnall, Simultaneous acquisition of multiple resolution images for dynamic contrast enhanced imaging of the breast, Mag. Res. in Med. 46:503-509, 2001.
3. J.V. Hajnal, D.L.G. Hill, D.J. Hawkes (eds.), Medical Image Registration, CRC Press, 2001.
4. F. Maes, A. Collignon, D. Vandermeulen, G. Marchal, P. Suetens, Multimodality image registration by maximization of mutual information, IEEE Trans. Medical Imaging 16(2):187-198, 1997.
5. J.-P. Thirion, Image matching as a diffusion process: an analogy with Maxwell's demons, Medical Image Analysis 2(3):243-260, 1998.