Y. Ogura, K. Utsunomiya, I. Narabayashi, T. Komori, I. Adachi, A. Stundžia*, P. Dufort*, M. Ichise**.
Osaka Medical College, Takatsuki, Japan;
*Advanced Biologic Corp., Toronto, Canada;
**National Institute of Mental Health, Bethesda, USA
Objectives: We developed a rapid non-linear technique to register and fuse 3D lung CT images with SPECT 99mTc MAA perfusion, 81mKr ventilation and ventilation–perfusion normalized (V/Q) images called the LungGuide Image Fusion Method (LG-IFM).
Methods:
CT, SPECT
99mTc-MAA perfusion, and 81mKr
ventilation lung imaging
was performed for 12 patients with suspected lung defects including embolism,
emphysema, and tumour nodules. For each patient, the
LG-IFM technique was applied in four stages. First, a plane of approximate
left-right symmetry, labeled as the midsagittal (MS) plane, separating the left
lung from the right lung, was found. This MS plane identification was performed
independently for CT and SPECT image data. Second, the
optimal linear match between the original lung SPECT and CT images was found
by aligning the MS planes and then rigidly translating (shifting) and rotating
the SPECT image in 3D such that the maximum overlap between the two images was
achieved, as measured by a statistical correlation function. Third, the most
probable SPECT outer lung surface was estimated and corresponding anatomic reference
points on the CT and SPECT lung surfaces were respectively identified. A one-to-one
correspondence between all points on the CT lung surface and all points on the
SPECT lung surface was generated giving a continuous 3D vector
displacement field 
from one surface to the other. Finally, a scalar deformation energy function
was constructed for a volume displacement field 
defined on the interior points of the lungs, with the surface mapping 
as a boundary constraint. Minimization of the energy function yielded a non-linear
mapping of the SPECT volume onto the CT volume, or equivalently 3D nonlinear
image fusion of CT with SPECT.
Results: The processing of same patient 3D LG-IFM registration and fusion of lung CT images (16 million voxels 32 MB) with SPECT ventilation, SPECT perfusion, and V/Q images (1 million voxels 2 MB) took 300 seconds on a PC workstation (Windows 2000 on a 2 GHz Pentium 4 with 512MB of RAM). The accuracy of image registration and fusion were confirmed visually for all twelve cases by 3 independent readers, using the lung defects and independent markers as reference.
Conclusion: The LungGuide Image Fusion Method appears a rapid and accurate method of generating fused 3D images of anatomic CT and functional SPECT lung studies within the same subject, and may be a clinically useful tool.