Asymmetry analysis using automatic segmentation and classification for breast cancer detection in thermograms

Thermal infrared imaging has shown effective results as a diagnostic tool in breast cancer detection. It can be used as a complementary to traditional mammography. Asymmetry analysis are usually used to help detect abnormalities. However, in infrared imaging, this cannot be done without human interference. This paper proposes an automatic approach to asymmetry analysis in thermograms. It includes automatic segmentation and pattern classification. Hough transform is used to extract the four feature curves that can uniquely segment the left and right breasts. The feature curves include the left and the right body boundary curves, and the two parabolic curves indicating the lower boundaries of the breasts. Upon segmentation, unsupervised learning technique is applied to classify each segmented pixel into certain number of clusters. Asymmetric abnormalities can then be identified based on pixel distribution within the same cluster. Both segmentation and classification results are shown on images captured from Elliott Mastology Center.