History of the Kurt Rossmann Laboratories

Kurt Rossmann, Ph.D., joined the faculty of the University of Chicago in 1967 as Professor and Director of the Section of Radiological Sciences, having worked earlier at the Kodak Research Laboratories.  Dr. Rossmann had pioneered modern x-ray imaging research by establishing the basic concepts of quantum mottle and modulation transfer function (MTF) in screen-film systems, and he continued to extend and apply those concepts at Chicago.  In 1969, Charles E. Metz, Ph.D., and Kunio Doi, Ph.D., joined his group to form the basis for the Center for Radiologic Image Research, which was funded by the National Institute of General Medical Sciences from 1972 to 1977.  During this time, C. E. Metz, Lee B. Lusted, M.D., and David J. Goodenough, Ph.D., (17, 27, 70) laid the foundation for use of receiver operating characteristic (ROC) analysis in the evaluation of radiologic images; Irving A. Lerch, Ph.D., and Arthur G. Haus (6, 25) made early measurements and analysis of x-ray spectra for optimization of technical factors related to angiography and mammography; and Carl J. Vyborny, M.D., Ph.D., (94, 95, 110) extended x-ray spectral analysis to measure the energy response of screen-film systems by means of a monoenergetic source using various fluorescent target materials.  Tragically, Dr. Rossmann died in 1976 at the age of 50.

In 1977, the Kurt Rossmann Laboratories for Radiologic Image Research were established in the Department of Radiology in memory of the late Professor Rossmann, with Kunio Doi as Professor and Director.  The central theme of research in the Rossmann Laboratories was, and remains, the improvement of diagnostic accuracy of radiologic imaging and minimization of patient exposure during radiographic examination.  The laboratory’s close-knit team of physicists and radiologists has been very productive, publishing more than 900 articles in the last three decades (1970-2004).  This work has been supported by grants from the National Institutes of Health, the American Cancer Society, The Whitaker Foundation, the Department of Energy, the U.S. Army and other Foundations, as well as generous support from numerous colleagues in industry around the world.

In the late 1970s, Harry K. Genant, M.D., and K. Doi (51, 68, 80) carried out extensive studies on high-resolution skeletal imaging techniques based on optical and radiographic magnification.  K. Doi and Eugene E. Duda, M.D., (75, 128) developed a stereoscopic angiography technique using the longitudinal magnification effect and a small focal spot shift, which became commonly used in some countries.  Basic imaging properties of rare-earth screen-film systems were investigated extensively by Gunnila Holje, Ph.D., and Yoshie Kodera, Ph.D., (138, 168, 195) and these systems subsequently were adopted for routine use in many radiology departments.  Heang-Ping Chan, Ph.D., (126, 134, 144) carried out extensive Monte Carlo simulation studies for the analysis of scattered radiation, absorbed dose, and anti-scatter grid performance in diagnostic radiology, which led to the development of high strip-density anti-scatter grids.  Leh-Nien Loo, Ph.D., working with K. Doi and C. E. Metz, (169) investigated the relationship between physical image quality indices and observer performance in the radiographic detection of simple test objects, and established the importance of signal detection theory in the optimization of observer performance in radiologic imaging.

During the 1980s, Maryellen L. Giger, Ph.D. (163, 164, 184) investigated fundamental imaging properties in digital radiography and provided the basis for determining and interpreting the MTFs, noise Wiener spectra and signal-to-noise ratios of digital radiographic systems by taking the digital sampling effects into account.  Extensive observer performance studies were carried out by Heber MacMahon, M.D., (207, 246, 318) and colleagues to investigate the effects of pixel size, image processing techniques, data compression, and CRT display on the detection of pulmonary lesions in digital chest images.  This series of observer studies has received widespread recognition as a landmark achievement.

Since the early 1980s, research effort in the Rossmann Laboratories has focused primarily on the development of computer-aided diagnostic (CAD) schemes for detection and differential diagnosis on radiological images.  At the initial phase of research and development, our CAD schemes included the detection and classifications of lesions in chest radiography, mammography, angiography, and bone radiography.  In chest radiography, computerized schemes were developed by M. L. Giger, Tsuneo Matsumoto, M.D., and others (235, 274, 350) for automated detection of nodules; by Shigehiko Katsuragawa, Ph.D., et al. (241, 257, 285), for detection of interstitial diseases; by Nobuyuki Nakamori, Ph.D., et al. (293, 325), for detection of cardiomegaly; and by Shigeru Sanada, Ph.D., et al. (329, 357) for detection of pneumothorax.  In mammography, Robert M. Nishikawa, Ph.D., and Robert A. Schmidt, M.D., (392, 394, 422) were carrying on work initiated by H. P. Chan (214, 228, 268) for automated detection of clustered microcalcifications, whereas methods for automated detection of breast masses were developed by M. L. Giger, Fang-Fang Yin, Ph.D., C. J. Vyborny, and Ulrich Bick, M.D. (275, 330, 443).  In angiography, Kenneth R. Hoffmann, Ph.D., Hiroshi Fujita, Ph.D., Noam Alperin, Ph.D., Laura E. Fencil M.D., Ph.D. and K. G. Chua, M.D., (204, 232, 283) were developing computerized schemes for quantitative analysis of stenotic lesions and blood flow rates, and determination of the 3D vascular tree.  C. E. Metz and L. E. Fencil (258, 272) developed a new method for determination of 3D structure in angiography without prior knowledge of the relationship between the two views.  In bone radiography, M. L. Giger and Philip Caligiuri, M.D., (366, 404) developed computerized methods for quantitative analysis of osteoporosis and risk of fracture.  Artificial neural networks (ANNs) were applied to the differential diagnosis of interstitial diseases by Naoki Asada, Ph.D., et al., (266) and to the task of distinguishing between benign and malignant lesions in mammograms by Yuzheng Wu, Ph.D. (398).  Methods for the computerized classification of clustered microcalcifications were developed by R. M. Nishikawa, R. A. Schmidt, and Yulei Jiang, Ph.D., (454) and of extracted mass lesions by M. L. Giger and Zhimin Huo, Ph.D., et al. (410, 506)

During the 1990s, further progress on CAD has been made in improving the performances of various CAD schemes.   In mammography, advanced CAD schemes for detection and classifications of masses and clustered microcalcifications were developed by two groups of investigators directed by M. L. Giger and R. M. Nishikawa, respectively, together with Z. Huo, Matthew A. Kupinski, Ph.D., Y. Jiang, and R. A. Schmidt et al.  For an improved CAD scheme for detection of clustered microcalcifications, Wei Zhang, Ph.D. et al. (440, 542) applied a shift-invariant ANN, Takehiro Ema, M.S. et al. (451) developed a new method for reduction of false positives based on local edge-gradient analysis, and Hiroyuki Yoshida, Ph.D. et al. (438, 534) applied the wavelet transform for multi-resolution analysis on microcalcifications.  Z. Huo et al. (602, 713, 762) further developed a method for classification of masses by analysis of spiculation.  Y. Jiang et al. (667, 763, 765) developed a CAD scheme for classification of clustered microcalcifications by use of ANN and extracted image features.  M. Kupinski et al. (609) developed a new method for automated seeded lesion segmentation on mammograms.  R. A. Schmidt and R. M. Nishikawa (428, 735) demonstrated the usefulness of CAD in detecting missed lesions in mammograms.  Z. Huo further developed mammographic image analysis for use in the computerized assessment of breast cancer risk. (709, 811)

In chest radiography, advanced CAD schemes were developed by a number of investigators including Xin-Wei Xu, Ph.D. et al. (474, 533, 583) on detection of lung nodules, Katsumi Nakamura, M.D., Masahito Aoyama, Ph.D., et al. (733, 791) on distinction between benign and malignant nodules, Takayuki Ishida, Ph.D., Junji Morishita, Ph.D. et al. (462, 566, 604) on detection of interstitial opacities, Kazuto Ashizawa, M.D., Hiroyuki Abe, M.D. et al. (633, 634, 790) on differential diagnosis of interstitial diseases, and Samuel G. Armato, III, Ph.D. et al. (403, 588, 589) on detection of gross abnormalities, segmentation of lung fields in PA and lateral views, and measurement of costophrenic angles.  In addition, Akiko Kano, Ph.D., and T. Ishida et al. (415, 664, 665) developed a new scheme for temporal subtraction of chest images for detection of interval changes, and Qiang Li, Ph.D. et al. (722, 725) developed a contralateral subtraction technique for detection of asymmetric abnormalities.  Junji Shiraishi, Ph.D., and S. Katsuragawa et al. (737) developed a publicly available database for digital chest images with lung nodules, which is currently being distributed by the Japanese Society of Radiological Technology.

For analysis of computed tomography (CT) images, M. L. Giger and K. T. Bae, M.D., Ph.D. et al. developed methods for lung nodule detection (409) and automated segmentation of 3-D liver structure. (364) S. G. Armato et al. (629, 746) further developed methods for detection of pulmonary nodules using 3D image features, and Martin Fiebich, Ph.D. et al. (557, 597, 653) developed segmentation techniques for vessels and bones in CT angiography.  Y. Samara, Ph.D. et al. (685, 686, 687) developed a CAD workstation of colon cancer, and Y. Masutani, Ph.D. et al. (727, 728) developed a CAD scheme for detection of pulmonary embolization.  Kenji Suzuki, Ph.D. et al. (890, 891) developed a massive training artificial neural network (MTANN) for a substantial reduction in the number of false positives in the detection of lung nodules in low-dose CT for lung cancer screening.  Q. Li et al. (874) developed 3 selective filters for significant enhancement of nodules, vessels, and airway walls.  Feng Li, M.D., et al. (871) investigated radiological, pathological and clinical features on missed lesions obtained with low-dose CT from a large-scale screening program for lung cancer, and S. G. Armato, and F. Li et al. (795) demonstrated that our CAD scheme can detect the majority of these missed lesions. 

Development of CAD methods was extended to breast imaging with magnetic resonance imaging (MRI) and ultrasonography. Kenneth G. A. Gilhuijs, Ph.D. et al. (600) developed an automated analysis method of breast lesions in three dimensions using dynamic MRI.  Karla Horsch, Ph.D., Karen Drukker, Ph.D., and M. L. Giger et al. (760, 799, 809) developed methods for the detection and characterization of breast lesions on ultrasound.  M. L. Giger et al. (707) developed an intelligent CAD workstation using similarity to known lesions and multiple visual prompt aids.  For automated detection of polyps in CT colonography, H. Yoshida and Janne Nappi, Ph.D. et al. (824, 841, 842) developed a new CAD scheme based on analysis of volumetric features.  C. Jiang, Ph.D., Michael Chinander, B.A., and M. L. Giger et al. (643, 666, 699) investigated computerized methods for characterizing bone patterns in skeletal radiographs.  Jacqueline Estappan, Ph.D., K. R. Hoffmann et al. (562) developed a computerized method for determination of 3D orientations of catheters from single projection images.  Anindya Sen, Ph.D., K. R. Hoffmann et al. (689) developed a method for tracking coronary vessels in angiograms.

Extensive studies on observer performance by use of ROC analysis with support by C. E. Metz have been carried out over the years to provide clear evidence that CAD can improve radiologists’ performance in the diagnosis of various diseases.  These studies included the detection of mammographic microcalcifications by H. P. Chan et al. (268) in 1990, the classification of microcalcifications by Y. Jiang et al. (607) in 1998 the classification of masses by Z. Huo et al. (811) in 2000, the detection of pulmonary nodules in chest images by Takeshi Kobayashi, M.D. et al. (511) in 1996, the classification of pulmonary nodules by J. Shiraishi et al. (832,887) in 2002, the detection of interstitial opacities by Lawrence Monnier-Cholley, M.D. et al. (613) in 1998, differential diagnoses of pulmonary interstitial diseases by K. Ashizawa et al. (633) in 1999, detection of interval changes by use of temporal subtraction of chest images by Matthew C. Difazio, M.D. et al. (552) in 1997, and detection of lung nodules by use of contralateral subtraction of chest images by Shunji Tsukuda M.D. et al. (833) in 2002.  In addition, results of a large scale observer test without and with CAD for detection of pulmonary nodules were reported by H. MacMahon et al. (677) in 1999.  Q. Li et al. (873) investigated new psychological measures for selecting similar images in low-dose CT for distinction between benign and malignant nodules, and demonstrated the usefulness of similar images in radiologists’ image interpretation in 2002.  These extensive observer studies were performed by a large number of radiologists who participated as observers; they are clinical staff in the Department of Radiology, the University of Chicago and also many clinical colleagues from many institutions around the world.  Detailed analysis of experimental data was carried out by use of new methods and algorithms related to ROC analysis developed by C. E. Metz et al. (209, 260) over the years.

In 1994, the first prototype intelligent CAD workstation was developed by a team led by R. M. Nishikawa (465, 521, 683) for computerized detection of masses and clustered microcalcifications on digitized mammograms, and was evaluated on over 25,000 screenees in the Department of Radiology at the University of Chicago.  One of the findings was that the computer was able to detect about 50% of “missed” breast cancers on previous mammograms.  Since 1996, we have presented the real-time interactive demonstration of various CAD workstations (844) together with ROC analysis in chest radiography and breast imaging in the Exhibit area at the Annual Meetings of the Radiological Society of North America (RSNA) in Chicago.  This interactive demonstration allowed numerous participants at the RSNA to gain personal experience on the concept of CAD and the benefits and limitations of CAD.  In 1996, we hosted the Third International Workshop (487) on Digital Mammography in Chicago.  In 1998, we also hosted the First International Workshop (645) on CAD in Chicago, and the subsequent Second (721), Third (768), Fourth (821); Fifth (870), and Sixth (924) CAD Workshops were held in conjunction with the 14^th, 15^th, 16^th, 17^th , and 18^th International Congress and Exhibition on Computer Assisted Radiology and Surgery (CARS) in San Francisco, 2000; Berlin, 2001, Paris, 2002; London, 2003; and Chicago, 2004, respectively.

New methods and new techniques related to CAD technologies were invented by more than 50 investigators in the Rossmann Laboratories, and have been licensed to companies including R2 Technology, Deus Technology, Median Technology, Mitsubishi Space Software Co., General Electric Corporation, and Toshiba Corporation.  The first commercial system for mammographic CAD was developed by R2 Technology, and received FDA approval for clinical use in 1998.  Since then, CAD for detection of breast lesions in mammograms has been used by many radiologists in clinical breast cancer screening programs.  The second commercial CAD system for detection of lung nodules in chest radiographs was developed by Deus Technology.  Mitsubishi Space Software Co. developed another CAD system by incorporating temporal subtraction on chest images.

Over the years, investigators in the Rossmann Laboratories have received numerous Honors and Awards for Scientific Exhibits at the RSNA, including two Magna Cum Laude Awards, two Excellence in Design Awards, many Cum Laude Awards and many Certificates of Merit.  In 1994, K. Doi received the Kurt Rossmann Memorial Award and gave the Inaugural Lecture for Memorial Lecture Series in Medical Physics, Upstate New York Chapter of the AAPM and the Eastman Kodak Company.  In 1995, the Stauffer Award for the best paper in Investigative Radiology was given to M. L. Giger, K. T. Bae and H. MacMahon (409) for “Computerized Detection of Pulmonary Nodules in Computed Tomography Images, and the Sylvia Sorkin Greenfield Award for the best paper in Medical Physics was given to P. Caliguiri, M. L. Giger and M. Favus (404) for “Multifractal Radiographic Analysis of Osteoporosis”.  In 1996, K. Doi was invited to give Eugene P. Pendergrass Annual Memorial Lecture at the Department of Radiology, University of Pennsylvania for his contribution to the development of computer-aided diagnosis.  In 1997, K. Doi received the Umetani Award from the Japanese Society of Radiological Technology for his contribution to imaging science and technology in diagnostic radiology.  In 1998, K. Doi received the Landauer Memorial Award from San Francisco Chapter of the AAPM.  In 2000, the Stauffer Award for the best paper in Academic Radiology was given to Y. Jiang, R. M. Nishikawa, R. M. Schmidt, C. E. Metz, M. L. Giger and K. Doi (667) for “Improving Breast Cancer Diagnosis with Computer-Aided Diagnosis”, and the Sylvia Sorkin Greenfield Award for the best paper in Medical Physics was given to T. Ishida, S. Katsuragawa, K. Nakamura, H. MacMahon and K. Doi (665) for “Iterative Image Warping Technique for Temporal Subtraction of Sequential Chest Radiographs to Detect Interval Change.”  In 2005, C. E. Metz will receive the 12^th Gray Medal from the International Comission on Radiation Units and Measurements (ICRU) at the World Congress of Medical Physics in Neurenburg, Germany, for his contribution to the development of ROC methodology in medical imaging.

In 2001, Diagnostic Imaging Magazine’s Excellence in Diagnostic Imaging Award in the category of Innovation was awarded to the Kurt Rossmann Laboratories for Radiological Image Research.  The January 2002 issue of Diagnostic Imaging states:  “The concept of computer-assisted diagnosis in radiology germinated at Kurt Rossmann Laboratories for Radiologic Imaging Research at the University of Chicago.  But the Rossmann Laboratories were more than just the point of origin of this concept; they transported CAD research to where it stands today.  CAD is gaining recognition as an important tool for improving diagnosis in tedious high-volume procedures or high-tech applications that involve huge data sets.  CAD is establishing a favorable reputation in screening mammography.  Software is being perfected to enable a similar role in chest radiography, low-dose CT lung cancer screening, colon cancer screening, and other areas.”

Many physicists who have participated in research projects at the Kurt Rossmann Laboratories were trained as students in the Department of Radiology’s Graduate Programs in Medical Physics, which are established in 2003 as the Committee on Medical Physics and offered jointly by the Departments of Radiology and Radiation & Cellular Oncology.  Graduate training in medical physics was established at The University of Chicago in 1955, with Professor Lester Skaggs, Ph.D., as Director, initially to provide training in Radiotherapy Physics and Health Physics.  The program was broadened to include the entire area of Medical Physics in 1967, when Kurt Rossmann joined the Department of Radiology.  The program was further expanded under the subsequent Directors:  C. E. Metz (1979-1985), K. Doi (1985-1998), and M. L. Giger (1998- present).  Since 1969, a total of 46 Ph.D. degrees and 26 S.M. degrees have been granted in the Physics of Diagnostic Radiology, Physics of Nuclear Medicine, and Physics of Radiation Therapy.  A strength of The University of Chicago’s Graduate Programs in Medical Physics is their focus on both diagnostic and therapeutic applications of medical image research.  Currently, 22 students are working toward the Ph.D. degree in the Programs, which are supported by a training grant from the National Cancer Institute, with M. L. Giger as Director.