Return to the Table of Contents.
Key words: computer-aided diagnosis, neural networks, image display, data compression, ROC analysis
This project is the continuation of various ongoing activities in performance assessment techniques (sensitivity, specificity, ROC curve, signal and noise analysis) in advanced areas of statistical pattern recognition, reconstructions from sparse data (fast imaging systems), signal detection and algorithmic observers, neural networks, data compression, chaos and fractal-based analysis, and computer-aided diagnosis. In FY 96, the emphasis has been on the establishment of a coherent regulatory approach to computer-aided diagnosis (CADx), bringing together the relevant device experiences of various Office of Device Evaluation (ODE) divisions and other Center components. A secondary goal was to advance the scientific basis for assessment of unconventional and artifact-limited imaging systems to the level previously achieved for conventional systems.
OST organized and moderated an open public workshop on computer-aided diagnosis which was attended by about 200 people from academia and industry. A summary was prepared and made available on the FDA World Wide Web site. The Office also contributed significantly to the review of computer-aided diagnosis premarket submissions for clinical laboratory and radiology devices.
OST scientists were three of the principal authors of the ICRU Report: Medical ImagingThe Assessment of Image Quality (ICRU Report #54, Bethesda, Maryland). This is the first major international document on the quantitative assessment of medical imaging systems.
OST purchased high-resolution, high-brightness image display hardware and upgraded software used to compare the performance of human observers with that of machine-implemented observers (e.g., conventional and neural discriminants) for detecting and discriminating lesions in noise- and artifact-limited images. The new software is written in a high-level language, making it portable between display devices. These upgrades extend the ability to assess the effect of limited-data-set reconstructions (e.g., high-speed imaging systems) on human and machine observer performance of lesion detection and discrimination tasks.
OST also coordinated a session on "Biomedical Applications of Neural Networks" at the World Congress on Neural Networks (WCNN - San Diego, California, 1996), and participated in a similar session at the International Conference on Neural Information Processing (ICONIP - Beijing, 1995). These sessions included clinical assessments of many forms of machine observers used for automatic lesion detection and discrimination in medical images. [PreME, ProA, Stds]
Return to the Table of Contents.
Key words: fluoride, bone strength, bone density, radiographic testing, mechanical testing
Various kinds of toxicity have been attributed to ingestion of fluoride, including dental fluorosis; bone fracture; reproductive, renal, gastrointestinal, and immunological toxicities; genotoxicity; and carcinogenicity. In 1990, a study by the PHS National Toxicology Program (NTP) found "equivocal evidence" of carcinogenicity associated with fluoride ingestion, based on a small number of osteosarcomas in male rats. In response to this finding, a subcommittee of the PHS Committee to Coordinate Environmental Health and Related Programs (CCEHRP) conducted a thorough review of the benefits and risks of water fluoridation and other sources of fluoride. The committee issued a report in 1991 recommending further study related to various regulatory requirements for fluoride. Regarding fluoride's effects on bone, CCEHRP recommended research to (1) determine the risk factors associated with the development of fluoride-associated osteosarcoma, and (2) further elucidate the mechanisms of fluoride action on bone at the molecular and physical chemical levels.
A collaborative study between CFSAN and CDRH was undertaken in response to the CCEHRP recommendations. The objective was to determine the levels of fluoride in the plasma, urine, and bone of rats at various stages of development and to characterize the bone strength for those rats. For each category of rat (male/female and pregnant/not pregnant at various stagesneonatal, weanling, and adult), five fluoride levels were studied (0 to 250 ppm in drinking water). CFSAN determined the fluoride content of one femur from each rat and CDRH conducted radiographic and mechanical testing on the contralateral femora to assess the effects of treatment on bone strength. For the radiographic imaging part of the study, computerized tomography testing was performed to measure bone density in the mid-diaphyseal and distal sections of the bone. This enabled scientists to study the effects of treatment on both cortical and trabecular bone. Plain-film radiographs of each bone were also made for a collaborative study with the University of Chicago using fractal analysis to estimate bone strength. Results from these imaging studies will be compared with results from mechanical (bending) tests on the bones to determine (1) the developmental effects of fluoride on bone strength, and (2) the correlation between bone density, fractal dimension and bone strength. DMMS and DECS, divisions within OST, have completed testing, and the data are being analyzed.
This study will provide FDA (and possibly EPA) with additional data to address regulatory concerns identified in the CCEHRP report on the benefits and risks of fluoride. The part of the study examining correlations between mechanical and radiographic test data may be useful to CDRH, CDER, CBER, and perhaps others with respect to evaluating imaging devices used to predict fracture risk, bone substitute materials, drugs used to treat or prevent osteoporosis, and methods for storing bone and bone marrow for transplant. [ProA]
Return to the Table of Contents.
Key words: mammography, x-ray spectra, clinical trial, ionization chambers, spectroscopy, automatic exposure control, rhodium anode x-ray tube
The Medical Imaging and Computer Applications Branch in OST conducts an ongoing program of research and evaluation in the general area of x-ray mammography. The program includes maintenance and operation of an experimental x-ray laboratory equipped with x-ray hardware typical of current clinical systems, and development of digital data and programs for the computer simulation of various aspects of mammography. These activities enable the branch to provide support for Center programs in premarket evaluation, compliance, and the implementation of the Mammography Quality Standards Act. Branch staff members also collaborate with scientists at Georgetown University, the University of Southern California, and the Diagnostic Radiology Department at the Clinical Center, National Institutes of Health (NIH) to study methods for improving the performance of x-ray mammography.
A major part of the effort expended on this project in FY 96 involved the optimized mammography system developed by CDRH in collaboration with the University of Southern California and the Clinical Center at NIH. The optimized system was used to study the feasibility of implementing digital mammography using photo-stimuable phosphor technology, sometimes known as computed radiography (CR). Figure 1, which is a plot of threshold contrast for a 200 micron disk target vs. x-ray fluence, shows the excellent latitude of the digital system relative to a conventional system. In addition to developing information on the possible role of CR technology in the evolution of digital mammography, this activity helped develop staff expertise in the evaluation of digital mammography systems in general. The activity was carried out in collaboration with colleagues at Georgetown University, who have access to the latest CR technology and considerable experience in its use. At the end of the reporting period, a first draft of a protocol for the clinical evaluation of the optimized system (with screen-film image receptor) was completed.
A substantial amount of effort was put into support of intramural mammography activities in other CDRH offices. Several presubmission meetings with manufacturers of digital mammography systems were attended, and a premarket submission for a computer-aided diagnosis system for mammography was reviewed. Consultation on MQSA proposed final regulations was provided. Spectroscopic measurements were done for the Radiation Metrology Branch to support development of the CDRH calibration facility for ionization chambers used to evaluate mammographic x-ray equipment during facility inspections, as well as the development of the NIST-designed crystal spectrometer. The crystal spectrometer measurements will help to extend its application from measurement of kVp to actual determination of quantitative spectra that may have additional applications in quality assurance and acceptance testing. [PreME, PostMS, Enf, Stds]
Return to the Table of Contents.
Key words: ultrasound, magnetic resonance spectroscopy, tissue characterization, liver disease, signal processing, silicone, bone densitometry
OST continues its program on the application of quantitative methods for tissue characterization using diagnostic ultrasound and magnetic resonance spectroscopy (MRS). This project is based on previously completed work on the fundamental physics of ultrasound imaging. It involves the extraction of "features" from ultrasound images and magnetic resonance spectra, and the combination of these features, to provide maximal diagnostic power. This work is intended to yield improved diagnostic capability for specific conditions. This work should also answer modality-independent questions about the statistical problems involved in decision-making based on multiple data inputs. A thorough understanding of these statistical issues is important for the premarket review of devices that provide diagnostic indications based on multiple features.
A clinical protocol at Georgetown University Medical Center is being followed to collect MRS and ultrasound data from patients with diffuse liver disease, including those with AIDS-related complications. Multimodality data from 20 normals and 10 abnormals were presented at the annual meeting of the Radiological Society of North America. This paper concluded that the metabolic information contained in the MRS data complements the structural information extracted from the ultrasound data, yielding improved differentiation of normal from abnormal liver.
OST is making strides in the development and assessment of novel signal processing methods in ultrasound and MRS. A paper on a method of estimation of ultrasonic scatterer number density has been submitted to the Journal of the Acoustical Society of America. This method was validated using ultrasonic data from tissue-mimicking phantoms.
More recently, OST has been investigating a new method for improved two-dimensional MRS localization and has demonstrated improved spatial resolution of the FFT for certain phantom data sets and in human liver in vivo. This method was shown to suppress contamination of liver signal by signal from muscle in magnetic resonance spectral maps. A paper on this topic has been submitted to the IEEE Transactions on Medical Imaging.
Another area of investigation is the use of MRS to detect the migration of silicone from breast implants to the liver. In collaboration with Georgetown University Medical Center, data will be collected on women with and without symptoms alleged to be caused by exposure to silicone. The information gained has the potential for aiding our understanding of the relationship between silicone and disease. The research effort to date has focused on the design of an unbiased patient enrollment procedure and the development of the MRS acquisition method. The MRS method has been developed and tested to show that it has adequate spatial localization and minimal interference from fat and water signals.
Two additional areas of investigation are ultrasound mammography and ultrasound bone densitometry. Data are being acquired for both of these applications under clinical protocols at Georgetown University Medical Center. To date, ultrasound data from 30 os calci (heel bones) have been acquired. Companion CT data have been acquired from eight of these subjects. A high correlation (r=0.85) between bone density assessed using CT and ultrasonic backscatter has been observed as shown in figure 2. This indicates potential for ultrasound to become a low-cost, nonionizing, portable alternative to CT for measuring bone density.
These efforts in magnetic resonance imaging and spectroscopy help build CDRH expertise necessary for related regulatory activities such as the review of 510(k)s and IDEs regarding highfield MR systems, MRI accessories (e.g., surface coils), MR compatibility issues (e.g., aneurysm clips and pacemakers), and new MR imaging techniques (e.g., high-gradient techniques, diffusion imaging, and functional imaging), as well as participation in related standards development activities. Similarly, these research efforts in ultrasound help build CDRH expertise necessary for related regulatory activities regarding state-of-the-art ultrasonic tissue characterization devices, such as imaging machines and bone densitometers. [ProA, PreME]
Return to the Table of Contents.
Key words: detective quantum efficiency, fluoroscopy, interventional radiology, observer SNR, signal-to-noise ratio (SNR), x-ray filtration, x-ray image intensifier
The substantial increase in the use of x-ray fluoroscopy as a visualization tool for a wide range of diagnostic and therapeutic procedures has focused OST attention on a couple of different aspects of imaging system performance. The overall goal of this project continues to be developing methods for accurately measuring the imaging performance of x-ray fluoroscopy systems and their associated components. These measures can then be used to optimize the imaging process so that the maximum benefit (SNR, contrast, etc.) is obtained for the lowest cost (organ dose, skin exposure, etc.).
Several efforts have been and are underway to accomplish the overall goal described above. These efforts include laboratory studies: to investigate the patient dose-reduction performance of adding different x-ray filter materials to the fluoroscopic x-ray beam; and to calculate a fundamental measure of imaging performance, i.e., the noise equivalent quanta (NEQ) or, similarly, the detective quantum efficiency (DQE) for x-ray image intensifiers. Since practical considerations make the measurement of the parameters defining the NEQ difficult in the clinical environment, algorithmic realizations of specified observers have been used to determine the "observer SNR" from sets of images with and without a lesion-like test object. Current research is focusing on using algorithmic observers to characterize the imaging system's performance in detecting a specific test object and on validating the connection between the NEQ and the "observer SNR."
Accomplishments in FY 96 include participation in an extensive refresher course on the "Physical and Technical Aspects of Angiography and Interventional Radiology" at the 1995 Annual Meeting of the Radiological Society of North America (RSNA), with a publication in the syllabus of methods of radiation exposure reduction in fluoroscopy titled "X-Ray Spectral Considerations in Fluoroscopy." In addition, prototypes of phantoms for imaging performance evaluation and associated analysis software have been developed for eventual use in the clinical environment. There is an ongoing research effort to compare the image scores from these phantoms to predictions derived from the fundamental measures obtained on two x-ray image intensifiers in Center laboratories. This research project also continues to provide the technical basis for several regulatory activities within the Center in the fluoroscopy area, including input into the drafting of guidance for premarket submittals associated with new solid-state x-ray imagers.
As mentioned above, there has been a substantial increase in the use of x-ray fluoroscopy as a visualization tool for a wide range of diagnostic and therapeutic procedures. The devices such as catheters and electrical leads used in these procedures have gotten smaller and smaller. In addition, there have been several occasions where x-ray fluoroscopy has been used as the screening tool for a component of a device recall program. These recalls have included detection tasks such as a broken strut on an artificial heart valve or the implanted electrical leads from an external pacemaker. The physical sizes of these devices approach a few tenths of a millimeter. For this reason, our initial effort in the digital x-ray fluoroscopy area is focusing on the effect of pixel size on SNR for digital x-ray fluoroscopy systems. A laboratory representation of an x-ray fluoroscopy system, which has the same characteristics as a system used for cardiac imaging in terms of physical geometry, x-ray beam quality, tissue-simulating phantom, radiation exposure per frame, x-ray image intensifier tube and CCD camera, has been used to generate the data for this initial effort.
The probability that a human observer will correctly detect a signal superimposed in fluoroscopic noise can be related to a signal-to-noise ratio (SNRSD) obtained from statistical decision theory developed by Markku Tapiovaara and Robert Wagner. The theory has provided a bridge between a rigorous and objective measure such as the NEQ and a subjective measure such as imaging phantom results through the use of algorithmic observers closely related to the human observer. These observers have been shown to correlate well with human observer performance.
In this preliminary analysis, we have calculated the digital signal spectrum for sharp, square objects of various sizes, SD(u,v), with the center of the object coinciding with the sampling grid. The digital Wiener spectrum was calculated using noise power measurements from our laboratory x-ray fluoroscopy system. These data were combined to calculate the "DC suppressed observer SNR (SNRDCS)." Figure 3 shows the results of these calculations for two different pixel widths and acquisition matrix sizes, i.e., 10242, 0.192 mm and 5122, 0.384 mm, respectively. The figure indicates the excellent agreement between the SNR obtained from fundamental measures such as the Wiener spectrum and the output of an algorithmic realization of the "DC suppressed" observer. (The uncertainty due to the finite number of samples is indicated on only some of the data points.)
However, of more interest in a public health sense, is the change in the SNRDCS as a function of object size. As is evident in the figure, there is a sharp drop in SNRDCS when the object size starts to approach dimensions equal to twice the pixel width. In those circumstances, one way to circumvent the loss of SNRDCSis to increase the number of x-ray photons. This translates directly to an increase in the radiation exposure to the patient. [PreME, Enf, ProA, PostMS]
Return to the Table of Contents.
Key words: computed tomography, diagnostic radiology, dosimetry, measurement, ionizing radiation
Manufacturers of computed tomography (CT) systems must provide dose data as part of the required user information prescribed in the CT section of the "Federal Performance Standard for Diagnostic X-Ray Equipment" (Code of Federal Regulations: 21 CFR 1020.33; 1984). The computed tomography dose index (CTDI), which is the multiple scan average dose (MSAD) at the center of a series of 14 contiguous scans, is the dose descriptor used in the Federal Performance Standard. The most direct way of getting the value of the MSAD in computed tomography is to employ a pencil-shaped ionization chamber for integration of a single-scan dose profile. Because the active length of the pencil chamber is fixed, the measurement can represent the value of the MSAD from a different number of contiguous scans depending on the slice thickness. This characteristic makes it difficult to compare the value of MSAD using the pencil chamber to the information required by Federal regulations on the computed tomography dose index (CTDI). For this reason, two alternative methods were developed by Center scientists to make the CTDI measurements at the center of a CT dosimetry phantom.
Accomplishments in FY 96 include the publication of a paper describing alternative methods of making CTDI measurements at the center of a CT dosimetry phantom. One alternative method involved the use of radio-opaque sleeves with the pencil chamber to limit the length of the single scan dose profile incident on the pencil chamber. In addition, TLD data obtained from the same CT systems were also used to obtain a set of conversion factors for converting the results of a measurement with the pencil chamber without a radio-opaque sleeve to a value of the CTDI. Figure 4 shows a plot of the set of the conversion factors for obtaining the CTDI from pencil chamber measurements. These data represent 103 different measurements of the conversion factors from TLD measurements of the single-scan profile for slice thicknesses ranging from 1 mm to 10 mm. The spread in the conversion factor at each slice thickness in terms of plus or minus one standard deviation are provided only on some of the points for clarity.
These conversion factors can be used to calculate the CTDI from a measurement on a single-scan dose profile with a pencil chamber having the same length of integration to the one used in the surveys. These values of the conversion factors are only appropriate for obtaining the CTDI at the center of the CT dosimetry phantom. The different methods of obtaining the CTDI using the TLDs, the pencil chamber with radio-opaque sleeves, and the set of conversion factors are all in general agreement, i.e., less than 10% difference on the average. Both alternative methods provide an easy and direct way of obtaining the CTDI from pencil chamber measurements, and provide a useful and simple method for clinical facilities to verify manufacturer's data and monitor system performance. [Enf, PostMS]