UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 1 Breast Tumor Segmentation
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 2 Breast Tumor Segmentation Presentation Overview Background and problem description Previous work Our approach Results Conclusion
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 3 Breast Tumor Segmentation Background Ultrasonic strain imaging A strain image is a spatial map of local deformation that occurs because of an applied load Obtained by comparing a pre-compression image to a post- compression image Tumors are stiff they show up as dark areas
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 4 Breast Tumor Segmentation The Problem Quantify contrast between tumor and background Must define tumor region and background region
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 5 Breast Tumor Segmentation Previous work Parametric active contours, aka snakes Strain Image Manual Segmentation by Radiologist Snakes Segmentation Wu Liu et. al. Segmentation of Elastographic Images using a Coarse-to-Fine Active Contour Model. Ultrasound in Medicine and Biology. Publication in progress.
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 6 Breast Tumor Segmentation Our Approach 1.) Smoothing filter 2.) Threshold 3.) Multiple morphological processing steps
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 7 Breast Tumor Segmentation Step by Step Original imageAveraging filter Threshold Remove holes Opening, remove fingers Isolate tumor Reverse opening distortion Result
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 8 Breast Tumor Segmentation Tumor Finding Three options for finding tumor: user-supplied coordinates, manual input, and automatic tumor finding. Automatic tumor finding: 1) Find the distance of each pixel from a black (0) pixel 2) Mark the pixel farthest from a black pixel and closest to the center of the image as inside the tumor
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 9 Breast Tumor Segmentation Some Results
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 10 Breast Tumor Segmentation Limitations
UW Madison Digital Image Processing Adam Slater and Matt McCormick / 12/14/2005 / Page 11 Breast Tumor Segmentation Conclusions Advantages Very accurate and precise Robust Disadvantages Loops in MatLab slow