History and Lessons from FDA Regulation of Digital Radiology

1. History and Lessons from FDA Regulation of Digital Radiology Kyle J. Myers, Ph.D. Division of Imaging and Applied Mathematics OSEL/CDRH/FDA October 22, 2009

2. From film-based to digital radiological imaging systems 1890s Roentgen discovers x-rays (film) 1970s Rare-earth phosphor screens 1976 Medical device amendments 1990s Solid state flat panel detectors

3. Digital Radiography • 510(k) for all applications other than mammography • No new types of safety or effectiveness questions • PMA required for full-field digital mammography • For imaging entire breast at once • Not small detectors used for local, diagnostic views

4. Full-field digital mammography • FDA assumed device would be used for screening; barred diagnostic-only claim • Target population is ALL women over 50 • Sensitivity and specificity impact safety and effectiveness, with potentially large consequences to public health • Missed cancers will not be screened again for 1 yr • Increase in false positives comes with additional exams, biopsy, patient anxiety

5. Why full-field digital mammography raised new questions: • High demand for resolution • Impact of discrete image pixels on visibility of microcalcifications and lesion margins? • Dynamic range of digital >> film • Potential for higher doses (not “self-limiting” like film) • Large image formats • Image stitching? • Other artifacts? E.g., dead pixels, rows, blocks, etc.

6. PMA data requirements for full-field digital mammography • Laboratory measurements • Preclinical images (phantoms) • Clinical study to determine diagnostic performance for screening

7. Lab Performance Data • Detailed engineering description • Spatial resolution: Modulation Transfer Function • Noise analysis: Noise Power Spectrum • Sensitometry: Gray scale transfer • Defect characteristics (# dead pixels, etc.) • Repeated exposure tests • Image erasure, fading, charge traps, etc. • Subject of international standards and consensus

8. Image of spiculated mass at center of breast Exposure at detector close to optimum for film-based ( 11 mR) Comparing analog mammography to digital Digital Nyquist frequency Detective Quantum Efficiency Film-based

9. Imaging Phantoms • Used for premarket analysis as well as post-market quality assurance • Subjective evaluation is standard • More sophisticated tests under development • ACR/MAP detectability phantom • 6 fibers • 5 speck groups • 5 masses

10. Quantitative phantom evaluations: predicting human performance • CDMAM contrast detail phantom • 4 Alternative Forced Choice • Signal size and shape known • Location unknown – one of four corners Observer predictions from lab measurements Human performance °+

11. Automated reading of phantom images • Quantitative data in terms of model observer SNR2 for objects in ACR/MAP • Note relative magnitude of SNR2 (group 3 and group 2)

12. Phantom design considerations • Involvement of professional societies • Mimic relevant clinical structures • Fibers  spiculations • Specks  microcalcifications • Disks  masses • Need to be made in quantities, with reliability of structures for fair comparisons • Field moving toward automated phantom scoring • Next challenge: phantoms for 3D systems

13. Why paired image interpretations failed to demonstrate substantial equivalence • Side-by-side comparisons of film-based and digital images of same patients? • Differences in images because of variability in images from patient repositioning • Same would be true for film-based images of same patient • 2D image of 3D object: variation in realization of overlapping tissues • Agreement of paired clinical interpretations? • Approach failed because of reader variability

14. PMA approvals required studies of diagnostic accuracy • Reader studies using Multiple-reader Multiple-case ROC analysis • See ICRU Report #79: Receiver Operating Characteristic Analysis in Medical Imaging • Multiple readers to sample range of reader skill and aggressiveness (5 in first approval) • Multiple cases (44 cancers, >600 cases in first approval) to sample range of case difficulty • Masses of different sizes, microcalcifications • Allowed inclusion of some diagnostic cases (case enrichment) to reduce sponsor’s burden • only 2-4 cancers per 1000 in screening cohort • MRMC ROC accounts for reader variability and differences in reader threshold from use of case enrichment

15. 10 years later • Five full-field digital mammography systems approved since 1999, with many supplements for engineering updates • Large NCI-funded prospective trial data released (>40,000 patients) • Demonstrated equivalence or better performance of FFDM relative to film • Moving toward down-classification from Class III to Class II with Special Controls Pisano, NEJM, 2005

16. Expected class II paradigm • Lab measurements • Phantom data • Clinical data • Demonstrate adequate patient positioning • Visibility of clinical structures • Evaluate artifacts • Radiological Devices Panel on 11/17/09

17. Major components in modern digital medical imaging Image Acquisition Image Processing Image Display Reader Computer-aided Diagnosis Picture Archiving and Communication System (PACS) Accessories: separate premarket applications

18. Image Compression • Medical image storage and communications devices (892.2010 and 892.2020) • Class I exempt • May apply irreversible compression with labeling • PACS systems (892.2050) • Class II • May apply irreversible compression with labeling • Mammography Quality Standards Act prohibits lossy compression for primary interpretation • Lossy compression can be applied to mammogram used for comparison (prior) • Primary read must use full data set

19. Summary • It has taken ten years for experience with full-field digital mammography to accumulate to support reclassification • Special controls for digital mammography will consist of a combination of bench measurements, phantom images, and clinical data • Lack of standardization of acquisition systems, processing software, and display devices can present issues