FY 2000 Statistical Issues Of Diagnostic Modalities
- Software development for Multivariate ROC Assessment of Diagnostic Modalities and Systems for Computer-Aided Diagnosis
- Tissue Characterization
This program addresses new and increasingly sophisticated computer techniques applied to medical diagnosis. These include processing of medical images or other medical data to search for signs of abnormality, with simple identification of suspicious regions for further review by a physician, or the offering of diagnostic confidence levels on abnormalities, or providing subtle quantitative physical or chemical information to the practitioner. These computerized medicine, tissue characterization, or computer-aided diagnosis (CADx) approaches use advanced statistical tools for diagnostic decision making under uncertainty, including classical Bayes' discriminants, neural-network architectures, and fuzzy logic. OST has played a significant role in device submissions that include those for automated Pap smear readers, lung cancer, and breast cancer detection devices. The computer applications program develops assessment methodologies for these systems based on the principles of statistical decision analysis.
Software development for Multivariate ROC Assessment of Diagnostic Modalities and Systems for Computer-Aided Diagnosis
Key words: computer-aided diagnosis, ROC, sensitivity, specificity
OST has previously developed a multivariate (here, six-component) statistical model for multiple-reader ROC studies of diagnostic imaging modalities. (The ROC, or receiver operating characteristic, is the graph of the trade-off between sensitivity and specificity of a diagnostic modality.) This model was applied to the design and analysis of clinical studies used during the approval process for digital mammography to account for patient and reader variability, their interaction, and the correlation of these components across modalities, conventional and digital.
The emerging contemporary problem of comparing unaided readers (radiologists) with readers who use a computer-assist modality requires a more elaborate treatment. This year OST developed more general nine-component ROC models to accommodate this more general task. OST scientists have now analyzed a number of important academic data sets using these models and can see to what extent the computer assist modifies the reader components of variance. These new tools provide a quantitative approach to assessing diagnostic imaging modalities, computer-assists for image readers, and training radiologists. They provide the ability to design a large pivotal trial from a smaller pilot study. And for the first time, they open up an approach to modeling the marginal effects of changes in all of the above reader and computer effects as well as changes in the physical characteristics of new imaging technologies. The tools were used in analyzing a PMA for computer-assisted lung cancer detection on chest images. The adoption of these tools in the imaging community will lead to more efficient trial designs for all kinds of medical imaging modalities.
Key words: ultrasound, magnetic resonance imaging, spectroscopy, tissue characterization
This project is the application of quantitative methods for tissue characterization using ultrasound and magnetic resonance. An understanding of the physics of these modalities and the statistical issues involved in multi-parameter tissue characterization is important for reviewing diagnostic devices. OST continued to work with the American Institute of Ultrasound in Medicine Technical Standards Committee Working Group on Backscatter Measurements to develop a standard for ultrasonic backscatter measurements in tissue. OST scientists continued their participation in developing ASTM "Standard Test Method for Evaluation of MR Image Artifacts from Passive Implants" and AIUM (American Institute of Ultrasound in Medicine) "Performance of Ultrasonic Backscatter Measurements." They continued to collect ultrasound and pathology data from ex vivo prostates at the University of Vermont to assess the usefulness of combining clinical, sonographic, and elastographic features to improve the detection of prostate cancer. OST also developed software to compute backscatter coefficient, texture, and elastographic parameters from prostate samples.