Margrit Betke CS585 Project Team Boston University Computer Science Margrit Betke PhD Harrison Hong MA William Mullally MA Chekema Prince BA Deborah Thomas BA Jingbin Wang ME New York University Medical School ID: 999979
Download Presentation The PPT/PDF document "Diagnostic Image Analysis of Chest Compu..." 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. Diagnostic Image Analysis of Chest Computed Tomography ScansMargrit BetkeCS585
2. Project TeamBoston University Computer Science:Margrit Betke, PhD Harrison Hong, MAWilliam Mullally, MA Chekema Prince, BA Deborah Thomas, BAJingbin Wang, MENew York University Medical School: Jane P. Ko, MD
3. Clinical Motivation EmphysemaAsthmaMetastasis of breast, prostate, colon cancer, melanoma, etcLung cancer
4. Metastatic Disease8.2 million people with a history of cancer in the US.Chest Computed Tomography (CT):Diagnose pulmonary metastasis of oncology patients.Evaluate response to treatment regimens.Repeated CT studies: Determine growth rates of pulmonary nodules.
5. Lung Cancer Screening“New” low-dose helical CT scan technology:Screen patients at high risk for primary lung cancer ?Lung cancer kills 160,000 people in the US per year.5-year survival rate of lung cancer patients: 15%Early detection and resection at Stage I : 5-year survival rate: 70%
6.
7.
8. Partial Volume Effect
9. Long-term Research GoalAutomated, quantitative, and efficient image analysis system to support radiologists in evaluating chest CT scans.System may improve patients' prognosis.
10. Research in my group Chest landmark detectionThorax, lung, fissure, trachea, and nodule segmentationNodule shape analysisRegistration of lung surfaces, nodules, vesselsNodule detection and classificationPhantom studies for validation purposes
11. Landmark DetectionTrachea, CarinaSternumSpineDetect landmarks using correlation of online templatesS (I – E[I])(q – E[q]) / (sI sq)
12. Lung Segmentation
13. Contour Smoothing
14.
15.
16. 3D Connectivity
17. 3D Lung Surfaces
18. 3D Lung Surfaces
19.
20.
21. Anterior Junction Line Problem
22. Solution to Anterior Junction Line Problem
23. Find Structures within LungConvert to binary imageFind connected componentsLabel connected components
24. Identifying Nodules
25. Classification of Lung Structures pink:noduleblue:vessel
26. Circularity Measure
27. E = (a+b) - (a-c) cos2q - sin2qShape: Round or Elongated?Axis of least inertia:Direction of elongation = Line for which sum of square of distance to points in object is a minimumE = SS r 0<=Emin/Emax<=1xyxr2212121b + (a-c)sin2q =b 22+-b + (a-c)cos2q =a-c 22+-q
28. Classification via Horizontal Regions
29. Classification Rules2D axial areaDistance to lung centroidsDistance to lung borderShape:Elongated: vessels, round: nodulesEvaluate shape by computingRatio of axes of second momentsPercentage of pixels within circumscribing circle
30. Nodule Detection Results318 / 370Radiology January 2001
31. Data Science Bowl 2017:Can you improve lung cancer detection?Data: > 1,000 CT scansGround truth: Will this patient be diagnosed with lung cancer within one year of the date of the scan? Yes or No?Competition 1st stage: training and testing data, 2nd stage: additional test dataBU team from AI class: ranked in 300s