Guest Editorial Fully 3-D Reconstruction of Medical Images

T ODAY, the majority of biomedical tomography systems on the market (e.g., modern positron emission tomog-raphy (PET), single photon emission computed tomography (SPECT), and X-ray cone-beam computed tomography (CT) devices) acquire projection data in 3-D mode. The detectors these devices employ have an axial extent which seems to increase steadily with the introduction of new systems. This allows a much more rapid collection of data (e.g., photons) than is possible with 2-D methods. The 3-D acquisition is exploited to 1) define faster scanning protocols, 2) obtain motion-correlated or dynamic (4-D) images, 3) obtain higher quality images, or 4) obtain images with lower radiation dose. This is because a large fraction of gamma or X-ray photons can be detected in multiring PET or in SPECT with large field-of-view detectors that otherwise would be lost. Three-Dimensional Acquisition: In order to optimally exploit the information that is contained in 3-D acquired data dedicated fully 3-D image reconstruction methods are necessary. In fact, the availability of fully 3-D reconstruction methods was a prerequisite for the introduction of 3-D tomography devices such as multirow CT scanners, flat panel detector CT scanners, multiring PET devices, and SPECT with, e.g., pinhole and cone-beam collimators. These systems have resulted in 1) many new clinical applications, such as rapid (total body) and new dynamic or cardiac imaging protocols with PET, SPECT, and CT, and 2) ultra-high resolution SPECT and PET and micro-CT of small laboratory animals, human organs (heart, breast, thyroid, joints), or tumors. The availability of fully 3-D reconstruction methods has also facilitated improved reconstruction from projections that traditionally, for the sake of simplicity and prac-ticality, was assumed to have a 2-D nature. For instance, in SPECT imaging with parallel-hole or fan-beam collimators, optimal compensation for the crosstalk between transaxial slices caused by detector blurring, collimator blurring, and photon scattering, can best be achieved using a fully 3-D reconstruction approach. This has led to strongly improved noise and resolution properties of present commercial systems. Imagining that the patient volume one wants to reconstruct consists of a stack of thin parallel slices, truly 3-D acquisition can be defined as follows. In 3-D acquisition, projection data of the 3-D distribution of properties in an object volume are obtained in such a way that at least a subset of projection pixels contains information about relevant properties of multiple slices, for all possible choices for slicing orientation. An example of such a …