Post

Computer-Assisted Detection Devices in Radiology: Performance Testing, Documentation and Labeling Requirements

This guidance applies to computer-assisted detection (CADe) devices applied to radiology images and radiology device data, classified under 21 CFR 892.2050. These devices are computerized systems intended to identify, mark, highlight, or direct attention to portions of radiology images or data that may reveal abnormalities during interpretation by clinicians. The guidance covers CADe devices marketed as complete packages with review workstations or as add-on software for imaging equipment or platforms.

  1. Determine if clinical performance assessment is needed based on ability to directly compare standalone performance with predicate
  2. Design comprehensive standalone performance testing protocol with independent test dataset
  3. Establish reference standard definition and scoring process before testing
  4. Develop detailed algorithm documentation including design, features, and processing steps
  5. Conduct generalizability testing across different imaging technologies
  6. Create user training program covering device capabilities and limitations
  7. Prepare comprehensive labeling including performance data and warnings
  8. Document all testing methodologies and maintain data integrity controls
  9. Submit electronic data for statistical analyses when possible
  10. Establish audit trail if reusing any test data

Key Considerations

Clinical testing

  • Clinical performance assessment may be required when standalone performance cannot be directly compared to predicate device
  • Clinical study should compare device performance to control modality (typically unaided reading)
  • Study should demonstrate statistical significance in performance improvement
  • Study population should be representative of intended use population

Non-clinical testing

  • Standalone performance testing required for all submissions
  • Testing database must be independent from training data
  • Performance metrics should include sensitivity and false positive rates with confidence intervals
  • Stratified analysis by relevant confounders required
  • Generalizability testing across different acquisition technologies needed

Human Factors

  • User training procedures must be provided
  • Training should cover device advantages and limitations
  • Training should help users identify appropriate device settings and reading scenarios

Software

  • Documentation level is generally “Basic” for CADe devices
  • Software documentation should follow FDA software guidance
  • Algorithm design and function must be described in detail
  • Processing steps, features, models and classifiers must be documented

Labeling

  • Must include indications for use, directions, warnings and precautions
  • Should describe device limitations and potential adverse events
  • Must include summary of clinical and standalone performance
  • Should specify compatible devices and acquisition techniques

Safety

  • Warnings about not relying solely on CADe output
  • Discussion of potential adverse events from false positives and missed abnormalities required

Other considerations

  • DICOM Std: Digital Imaging and Communications in Medicine Standard
  • JPEG Std: Joint Photographic Experts Group Standard
  • SMPTE Test Pattern: Society of Motion Picture and Television Engineers Test Pattern

Original guidance

  • Computer-Assisted Detection Devices in Radiology: Performance Testing, Documentation and Labeling Requirements
  • HTML / PDF
  • Issue date: 2022-09-28
  • Last changed date: 2022-09-27
  • Status: FINAL
  • Official FDA topics: Medical Devices, Radiation-Emitting Products, Radiology, Premarket
  • ReguVirta summary file ID: 5b6f4cc7c6f81c81ef864a4c9e419bcf
This post is licensed under CC BY 4.0 by the author.