Medical Image Processing - 1984 to 1999
Volume 13 from the series Selected SPIE Papers on CD-ROM
(available as an electronic publication from SPIE)
Introduction
SPIE Conferences have long been an incubator for new ideas in imaging and optical technologies. A huge number of new developments in medical imaging have been first publicized at SPIE conferences. In these notes I will provide an overview of the development of image processing in medical imaging, as represented in the SPIE press.
This collection of 377 SPIE papers on medical image processing predominantly covers the period from 1984 to 1999. It was in 1984 that the SPIE conference Applications of Optical Instrumentation in Medicine embarked on a new direction under the guidance of Roger Schneider and Sam Dwyer III. They reconstituted the conference series to focus on the technical aspects of medical imaging. Many of us provided moral support for this new direction and for that we were honored by being included on the Program Committee. In their introduction to the Proceedings for that year [Proc. SPIE, vol. 454], Roger and Sam outlined their intentions for the new series. They wrote that the "goal was to address the physics and statistics of image encoding by modality and the processing, display, archiving, management, and psychophysical considerations independently of modality".
I have included a number of papers from before 1984. Many of them are from the first SPIE conference on medical imaging in 1972, which was called Applications of Optical Instrumentation in Medicine. (Henceforth, I will refer to these conferences as simply Optical Instrumentation in Medicine.) These papers offer insight into the history of medical imaging. One finds the same concerns about the usefulness of image processing that we hear today. There were concerns about image quality and how to measure it, better reconstruction algorithms, optimization of imaging systems, etc. These are enduring issues, as they should be. The papers from the 1972 conference are also of interest because most of the authors in that volume have gone on to become leaders in the medical imaging field. In those days the Proceedings had a more personal touch; they included photographs of the authors.
One of the principal concerns at the 1984 conference was image quality and how to quantify it, which distinguished it from most other image processing conferences. I have accordingly chosen to include several of the foundational papers from that conference. These papers set the tone for the future. A few papers from the preceding years are also included to provide some background for the 1984 conference. There was a strong sentiment that an image’s quality should be related to how well visual tasks could be performed using that image. This task performance could be assessed in terms of signal-detection theory, as had been outlined by Bob Wagner [Photogr. Sci. Eng. 22, 41-46, (1978)], and others. Bob connected the signal-to-noise ratio for visual task performance by the ideal observer to the Noise-Equivalent Quanta (NEQ), which had been promoted as a quantitative measure of the image quality of an imaging system by Chris Dainty and Rodney Shaw [Image Science (Academic, 1974)].
Since 1984 the Medical Imaging series has undergone several metamorphoses. A few years later, the name of the conference was made even more unwieldy when it was changed to Applications of Optical Instrumentation in Medicine XIX and Picture Archiving and Communication Systems (PACS IV) for Medical Applications. In 1987 reason prevailed and the title became simply Medical Imaging. In 1989 the Medical Imaging Symposium was effectively born with the creation of four conferences: Image Formation, Image Capture and Display, Image Processing, and PACS System Design and Evaluation. By 1999, we had seven conferences, including Image Processing, Physics of Medical Imaging, Physiology and Function from Multidimensional Images, Image Perception and Performance, Image Display, PACS Design and Evaluation, and Ultrasonic Transducer Engineering.
As evidenced by the papers included on this CD, the growth of medical image processing from 1984 to 1999 has been phenomenal. In 1984 there were fewer than 25 papers on image processing. By 1999 there were almost 200 papers in the Image Processing conference alone, with many more papers devoted to a large extent to IP distributed among the other conferences within the Medical Imaging Symposium.
Beyond the fact that the number of medical-image-processing papers has mushroomed, there has been a tremendous maturation of the field. There is little doubt that the huge improvement in the sophistication of the techniques can be partially attributed to the immense increase in computing power that we have witnessed in the last decade or so. Better resources spurred researchers to think about more comprehensive models.
I have organized the papers in terms of categories, many of which reflect the organization of today’s Medical Imaging: Image Processing meeting. There is a table of contents for each category. In the following sections, I will provide a brief description of each paper. The papers in many categories have migrated over the years from the Image Processing conference to other conferences in the Medical Imaging Symposium. For lack of space, not lack of paper quality, I have tended to reduce the representation from these categories as they have moved into other conferences. Because some categories are related applications and others to techniques, many papers appear in more than one category.
The topic of Image Reconstruction has been well represented in the Image Processing conferences. It was the advent of computed tomography that paved the way for radiologists’ acceptance of medical images that had been processed by a computer. The CT modality required the computer to produce interpretable images from otherwise unintelligible measurements. Reconstruction algorithms have played a significant role in creating other new modalities, including MRI, PET, SPECT, and optical tomography. The acceptance of CT led to the question, are there other ways in which computers could help radiologists interpret medical images? That question naturally leads to another category highlighted here, Image Quality and Signal Detection Theory. As mentioned above, this topic was an important new thrust in the 1984 Optical Instrumentation in Medicine. Another prominent category is Geometric Models, which includes the important topic of deformable geometric models. Since 1990, deformable models have become prevalent in many areas of image analysis. Given their rate of growth, they are likely to become an essential tool in image processing.
Most of the papers included in this volume come from the Medical Imaging Symposium series. However, I have searched through the bulk of SPIE publications (the Proceedings alone account for nearly 100 meters of library shelf space) to find articles relevant to this volume. Around 10% of the papers contained herein are from over 20 conferences outside the Medical Imaging series.
I would like to thank many of the authors represented in this volume who provided helpful guidance in selecting the most representative medical-image-processing papers.
Acronyms commonly used in medical image processing
The following acronyms are commonly used in medical image processing: ANN, Artificial Neural Network; ART, Algebraic Reconstruction Technique; CAD, Computer Aided Diagnosis; CT, Computed Tomography; DQE, Detective Quantum Efficiency; DRS, Mayo Clinic’s Dynamic Spatial Reconstructor; DSA, Digital Subtraction Angiography; ECT, Emission Computed Tomography; EM, Expectation Maximization; IP, Image Processing: MAP, Maximum A Posteriori; MCMC, Markov Chain Monte Carlo; MEG, Magnetoencephalography; ML, Maximum Likelihood; MRF, Markov Random Field; MRI, Magnetic Resonance Image; NEQ, Noise-Equivalent Quanta; NN, Neural Network; NPS, Noise Power Spectrum; PET, Positron Emission Tomography; POCS, Projection Onto Convex Sets; ROC, Receiver Operating Characteristic; SNR, Signal-to-Noise Ratio; SPECT, Single Photon Emission Computed Tomography.
Description of this collection
This collection consists of over 350 papers on two CD-ROMs. The papers are organized in terms of the following categories, the history and importance of each of which is discussed in the editor’s introductory notes.