Duke University Jeffrey R Marks PhD Joseph Lo PhD Lars Grimm MD Moffitt Cancer Center John Heine PhD Erin Fowler MPH Emma Hume MPH Jared Weinfurtner MD Creatv MicroTech ChaMei Tang PhD Daniel L Adams BS ID: 1040501
Download Presentation The PPT/PDF document "Combining Automated Image Measurements, ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. Combining Automated Image Measurements, Blood Biomarkers, and Clinical Data for Improved Decisions in Diagnostic MammographyDuke University: Jeffrey R. Marks, PhD; Joseph Lo, PhD, Lars Grimm, MDMoffitt Cancer Center: John Heine, PhD; Erin Fowler, MPH; Emma Hume, MPH, Jared Weinfurtner, MDCreatv MicroTech: Cha-Mei Tang, PhD; Daniel L. Adams, BSNASA Jet Propulsion Laboratory: Luca Cinquini, PhD; Heather Kincaid; Ashish Mahabal, PhDFred Hutchinson Cancer Research Center: Data Management and Coordination Center
2. IntroductionBreast screening: 2% - 3% of women are classified as Breast Imaging Reporting and Data System (BI-RADS) 4 by the radiologist’s interpretationThese women are recommended for biopsy and about 80% are found breast cancer negativeConclude: computerized decisions could lead to improved specificity for this subset of women2Study OverviewCombine clinical data, blood markers (rare cell detection), and image measurements with AI (n = 1050 collected prospectively) for improved decisions Build EDRN repository with blood specimens, image data with expert annotation, clinical data, and pathological data
3. Truth File GenerationExpert annotation: each lesion is labeled individuallyAutomated Image analysis: directed to the abnormalityExample: mammogram with two abnormalities3
4. ProgressPatient accrual n ~ 914Rare cell detection in real timeImage annotationRepository constructionDevelop baseline models without automated image measurements or blood markers as a baseline performance comparison metrics4Clinical DataBlood Collection and Rare Cell DetectionImage Data
5. Baseline Modeling (masses only)5Model 1:Az = 0.89 (0.82, 0.95)MeasureOR (95% CI)Age2.10 (1.34, 3.28)Margin1.67 (1.11, 2.52)BI-RADS4.67 (2.55, 8.55)N = 265 Cancer = 47 Non-Cancer = 218Model 2:Az = 0.94 (0.89, 0.99)MeasureOR (95% CI)Age2.09 (1.26, 3.46)Margin2.01 (0.66, 6.09)BI-RADS1.40 (0.86, 2.29)Shape4.53 (2.33, 8.79)N = 226 Cancer = 37 Non-Cancer = 189
6. ConclusionSimple baseline modeling suggests more sophisticated techniques may lead to improved decisions clinicallyRepository development has many user-friendly attributesQuerying methodsDocumentationData structureImage file naming conventionsFinal modeling and unblinding will be developed later this year6