This document describes a framework for breast lesion segmentation in ultrasound images. It consists of pre-processing using median filtering to reduce speckle noise. Segmentation is done using normalized cut to identify four segments, which are then classified using k-means clustering. The best segmented lesion region is selected based on minimum area. The framework was tested on 14 images, achieving correct segmentation on 11 images based on Jaccard and Dice similarity metrics. While speckle noise reduction is important, the median filter was not effective for edge enhancement.
4. Introduction
Medical Imaging Analysis Module 4
Breast Lesion
Segmentation
Digital
Mammography
(DM)
Ultra-Sound
(US) imaging
Magnetic
Resonance
Image (MRI)
• Harmless and painless examination method
• Perfect early-stage cancer detection
• Reduce the potential number of unnecessary
biopsies
11. Framework: Classification
Medical Imaging Analysis Module 11
Norm
Input Image
One of
Segmented Images
kmeans
K-means Clustering
(2 clusters)
Contour Selection
Contour Selection with
Minimum Length
1st Approach
Main
2nd Approach
Backup
Contour Selection
Otsu Binary
Thresholding
Selection of
Best Classified
Image based on
Minimum Area
Extract
Lesion
Region
16. Conclusion
Medical Imaging Analysis Module 16
Speckle noise reduction:
It is an important prerequisite , whatever ultrasound imaging
techniques is used for tissue characterization.
preprocessing step:
The median filter in the preprocessing step is not an effective method
to enhance the edges and lines in the images.
17. Bibliography
Medical Imaging Analysis Module 17
“Automated breast cancer detection and classification using
ultrasound images: A survey”, H.D. Cheng, J. Shan, W. Ju, Y. Guo,
and L. Zhang, Pattern Recognition, Volume 43, Issue 1, January
2010, Pages 299-317.
“Automated segmentation of breast lesions in ultrasound images”, X.
Liu, Z. M. Huo, and J. W. Zhang, IEEE Comput. Soc., Shanghai,
China, 2005, pp. 7433–7435.
“Image Segmentation with Normalized Cuts”, Jianbo Shi,
Department of Computer and Information Science, University of
Pennsylvania.