Quantitative reconstruction of radiopharmaceutical-uptake distribution within human chest and segmentation of reconstructed uptake regions-of- interest (ROIs) using single-photon emission computed tomography (SPECT) with Tc-99m and/or TI-2Ol labeled agents require simultaneous compensation for photon attenuation, scatter, and depth-dependent detector response and suppression of Poisson noise. This project continues the long-term objectives of the Principal Investigator's NIH First Award: (i) to develop three-dimensional (3D) reconstruction methods which compensate efficiently for attenuation, scatter, and detector-response variation and suppress effectively noise propagation; and (ii) to investigate automatic segmentation methods for accurate quantification of reconstructed ROIs.
The specific aims of this project are: (l) To develop simultaneous compensation techniques - A recursive ray-tracing will be used to compute efficiently attenuation factors; the Klein-Nishina formula will be employed to determine accurately scatter contribution and compared with the multiple energy-window compensation techniques; a depth-dependent convolution will be adapted to compensate effectively for detector- response variation. (2) To develop reconstruction and segmentation algorithms - A maximum a posteriori probability expectation-maximization (MAP-EM) approach will be investigated; its effectiveness of noise suppression and its stability of convergence by including an edge- preserving noise-smoothing a priori constraint will be studied. (3) To implement the reconstruction and segmentation algorithms - A unified MAP- EM algorithm incorporating the compensation techniques will be coded efficiently for parallel-, fan-, and cone-beam reconstructions; the object-specific attenuation map will be reconstructed from transmission scans of a three-head SPECT system using an external radioactive line source (where the truncation of fan-beam collimation will be effectively compensated by a MAP-EM approach with a priori known attenuation coefficients); an automatic MAP-EM segmentation will be investigated to improve the attenuation map and to facilitate the quantification of reconstructed ROIs. (4) To evaluate the accuracy of reconstruction and segmentation - Criteria, such as root-mean-square error, bias-variance graph, and miss-segment ratio, will be used to quantify the reconstructed images and segmented ROIs against their simulated and experimental 3D anthropomorphic phantoms of human thorax. The proposed unified quantitative methods can reconstruct accurately the uptake distribution in a l28(3) array in less than a half hour using a HP/730 desktop computer, and segment automatically the ROIs. The methods should have significant impact on our ability to diagnose the heart and lung diseases (such as coronary artery disease and pulmonary embolism) and to probe their metabolism.
