Associate Professor Faculty of Engineering Multimedia University - PowerPoint PPT Presentation

Mohammad Faizal Ahmad Fauzi, Ph.D. Associate Professor Faculty of Engineering Multimedia University

 Imaging Informatics  Imaging basics  Imaging modalities  PACS and its core functions  DICOM

20 mm Nodule

4 mm Nodule

5 mm Nodule

Reader 1 Reader 1

Reader 2 Reader 2

Reader 3 Reader 3

 Fatigue  Distraction  Emotional stress  Variation in reader  Satisfaction of Search

 Breast cancer is missed 10-30% …  by Expert Mammographers

 Sensitivity of radiologists in detecting breast cancer on mammograms can be improved by 15% through double reading .

 Computer-aided diagnosis: ◦ a diagnosis made by a physician using the output of a computerized system  Computerized system ◦ Automated image (or data) analysis

 Breast Cancer  Lung Cancer  Brain Cancer  Colon Cancer

Find Six Differences

Find Six Differences

• Solitary Pulmonary Nodules • Microcalcifications • Ground Glass Opacities • Masses

Malignant Benign

HR 2 (7/23/01) 5 10 15 20 25 30 35 40 5 10 15 20 25 30 35 40

1 2 3 4 5 6 7 8 9 10 11 12

10 20 30 40 50 60 10 20 30 40 50

 Organ segmentation  Candidate detection/segmentation  Feature Extraction  Classification/clustering

 Organ segmentation  Candidate detection/segmentation  Feature Extraction  Classification/clustering

 Segment Lung Regions within the CT slice  Detect left and right lungs

 Segmented lung region may exclude some nodules adjacent to pleura  Connect edge points of concave regions  Recover potential nodules adjacent to pleura

d 2 d 1 P 1 d e P 2

 Organ segmentation  Candidate detection/segmentation  Feature Extraction  Classification/clustering

Mammogram Image Pre-screening Potential Signals CNN Classifier Potential TP Signals Clustering Microcalcification Clusters

Mammogram Image Pre-screening Potential Signals CNN Classifier Potential TP Signals Clustering Microcalcification Clusters

 Identify high density regions within segmented lung regions  Segmentation by k-means clustering with two classes: ◦ nodule candidates ◦ lung region

Identification of Blood Vessels Thin long structure V-shaped True structure nodule

 Organ segmentation  Candidate detection/segmentation  Feature Extraction  Classification/clustering

 Thin long structures ◦ Major-to-minor axis ratio of a fitted ellipse  V-shaped structures ◦ Rectangularity

 Thin long structures a R tl  b a b  V-shaped structures Area of rectangle  R v Area of object

 Organ segmentation  Candidate detection/segmentation  Feature Extraction  Classification/clustering

FP ROI TP ROI

Mammogram Image Pre-screening Potential Signals CNN Classifier Potential TP Signals Clustering Microcalcification Clusters

{ 0: FP CNN Classifier 1: TP INPUT ROI

Mammogram Image Pre-screening Potential Signals CNN Classifier Potential TP Signals Clustering Microcalcification Clusters

 Image ◦ How to represent ◦ How to generate it  Imaging modalities ◦ How to integrate ◦ How to manage  Image Analysis ◦ Radiology ◦ Big picture