Three-dimensional medical imaging: algorithms and computer systems

Reviewer: Gregory M. Aharonian

Medical imaging has always been one of the more glamorous aspects of biomedical computing. To see a heart beating in real time, to see the brain consuming glucose while thinking, or for a mother to see her yet-to-be-born baby have always caught the eye and mind much more than X-ray diffraction patterns or gel plots. The authors have written an excellent non-mathematical review of the problem of three-dimensional medical imaging. Medical imaging is a complicated data collection and analysis task. The raw data are collected from the highest of high-tech instruments—magnetic resonance imaging (MRI) positron emission tomography (PET) machines—in large quantities, reduced to bits, stored in various three-dimensional data structures, manipulated, and fused before being graphically displayed. The authors focus on three-dimensional imaging and display. The first part of the paper deals with representing three-dimensional data and objects inside the imaging computer program, and in particular with the problem of handling contours, surfaces, and volumes. Next, the authors review medical imaging rendering operations, including boundary detection, digital dissection, histogram equalization, multiplanar registration, and tissue/image registration. General descriptions of these topics are given along with ample references. Given the large amounts of data involved, one needs to closely couple algorithms with dedicated hardware to obtain responsive system performances. The authors conclude with a review of eight specific medical imaging machines, both experimental (such as Kaufman's Cube Architecture) and commercial (such as Pixar and Vicom II). These machines reflect the three distinct approaches that seem to have been settled on: software-encoded algorithms on general-purpose computers, graphics accelerators adapted for three-dimensional medical imaging, and special-purpose computer systems with hardware encoding of algorithms. A table with other software/hardware implementations is given. Appendices treat a few rendering operations in more detail, including hidden surface removal, ray tracing, and image segmentation. The paper concludes with a bibliography containing over 400 entries. The authors have prepared a good non-mathematical review of the problems and challenges of three-dimensional medical imaging. It provides a good overview as well as ample references for more detailed information on specific topics. Many pictures and architectural diagrams further illustrate the comments in the text.