Asymmetry analysis of breast thermograms using BM3D technique and statistical texture features

Medical thermography plays a significant role in early detection of breast cancer. The abnormality detection using asymmetry analysis is complex due to low contrast, low signal to noise ratio and absence of clear edges. In this work, asymmetry analysis is carried out on denoised breast thermal images. Block matching and 3D filtering technique (BM3D) is adopted for noise removal. The breast tissues are extracted from background tissue by multiplying ground truth masks with denoised images. The midpoint of inframammary folds is identified to separate left and right regions from segmented images. Normal and abnormal groups are categorized based on the healthy and pathological conditions of the separated breast tissues. Second order features of co-occurrence matrix such as energy, entropy, contrast and difference of variance are extracted from denoised and raw images. Features from denoised images are found to be very effective in discriminating abnormalities present in breast tissues. Hence, it appears that the features extracted from denoised images can be used efficiently to identify the abnormality breast thermograms.