1 / 27

Clinical Description – Part I William Sacks, PhD, MD--ODE

R2 Technology, Inc. ImageChecker CT CAD Software System P030012 Radiological Devices Advisory Panel February 3, 2004. Clinical Description – Part I William Sacks, PhD, MD--ODE. Outline Character of device Clinical utility Instructions for use Issues new to this device.

delling-ull
Download Presentation

Clinical Description – Part I William Sacks, PhD, MD--ODE

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. R2 Technology, Inc. ImageChecker CT CAD Software SystemP030012 Radiological Devices Advisory Panel February 3, 2004 Clinical Description – Part I William Sacks, PhD, MD--ODE Sacks

  2. Outline • Character of device • Clinical utility • Instructions for use • Issues new to this device Sacks

  3. Character of device • For chest CT scans • For CTs done for any indication • Detects solid lung nodules • E.g., not ground glass opacities (GGOs) • Nodules between 4-30 mm • Computer-Aided Detection (CAD) Sacks

  4. Computer-Aided Detection(CAD)/Diagnosis(CADx) CAD CADx Sacks

  5. Clinical Utility • Many nodules missed in clinical practice: • other pathology distracts • hundreds of images per exam • CAD intended to reduce missed nodules Sacks

  6. Instructions for Use • Reader reviews films unaided first • CAD marks candidate nodules • Reader looks again in vicinity of marks • If CAD fails to mark a nodule judged actionable on unaided review, reader should retain initial judgment Sacks

  7. Issues new to this device • CAD target • Definition of “truth” • Unit of analysis • Endpoints Sacks

  8. CAD TARGET • Not malignant nodule • but • Actionable nodule Sacks

  9. DEFINITION OF “TRUTH” • Not biopsy (tissue histology) • but • Expert panel Sacks

  10. UNIT OF ANALYSIS • Statistical unit = person • further broken down into • Lung quadrant Sacks

  11. ENDPOINTS • Sensitivity and specificity (or FPR) of action recommendation • and/or MRMC ROC Sacks

  12. CLINICAL STUDY • 3 expert radiologists on panel to determine “truth” for each nodule • 15 other radiologists with range of experience, called “the readers” • 90 subjects (360 lung quadrants) • 15 readers used 100-point scale for confidence-in-actionability of each case Sacks

  13. CLINICAL STUDY • Presentation by • Nicholas Petrick, Ph.D. Sacks

  14. R2 Technology, Inc. ImageChecker CT CAD Software SystemP030012 Radiological Devices Advisory Panel February 3, 2004 Clinical Conclusions – Part II William Sacks, PhD, MD--ODE Sacks

  15. What does gain in Azof 0.02 (95% CI: 0.01, 0.04)mean clinically? Sacks

  16. When CAD is used according to instruction to retain all judgments of actionability even if unmarked by CAD... Sacks

  17. ... user always maintains or increases sensitivity, and always maintains or increases FPR, as well. Sacks

  18. Gain of 0.2 in AZ understates relative gain in sensitivity. Sacks

  19. ROC 1 ΔFPR Aided FNRU ΔSe =ΔFNR Unaided SeU Chance line (guessing) TPR 0 1 0 FPR Sacks

  20. ROC 1 Loss of Se only possible if instruction not followed Aided Realistic Unaided Chance line (guessing) TPR 0 1 0 FPR Sacks

  21. So any stat sig improvement in Az means an even greater relative gain in Se, without “undue” increase in FPR. Sacks

  22. Can we infer from improved user performance in clinical study ... Sacks

  23. ... that user will improve performance with CAD in clinical practice? Sacks

  24. For example, in actual clinical practice, the unaided Az could be lowered by failure to read first as one would normally read (i.e., with adequate vigilance). Sacks

  25. If this were to happen, then the aided Az could also be lower than current practice (in the complete absence of a CAD). Sacks

  26. Two rules, if followed by user, will prevent that from happening: • Always read unaided first, and as carefully as if you had no CAD. • Never back off from unaided judgment of actionability of a nodule if CAD fails to mark it. Sacks

  27. Summary • Data show stat sig improvement of Az of aided over unaided readings with CAD • If “Always” rule followed then aided curve will be higher than current practice • If “Never” rule followed, then gain in Az implies even greater relative improvement in Se of aided readings over current practice, without “undue” increase in FPR Sacks

More Related