Exploration of fibro-glandular region and breast density classification of digitised mammograms using least square support vector machine

Breast tissue density is one of the significant risk-marker for identification of breast cancer in early stage. In the proposed work, fibro-glandular region is explored and classification of breast density as dense and non-dense is performed. Image pre-processing is performed to improve the image quality followed by segmentation of breast region to obtain region of interest (RoI). For the obtained RoI, pseudo colouring is performed to improve image acuity accompanied by R-image extraction and post-processing to obtain fibro-glandular breast tissues. Area, histogram, fractal, grey-level co-occurrence matrix and grey-level run length matrix features are derived from both fibro-glandular and RoI regions and ratiometric value of features are computed. Further, mutual-information-based feature ranking algorithm is applied on the derived ratiometric values and the significant features are identified. These significant features when fed to least square-support vector machine produced average classification accuracy (%) of 86.1 ± 6.03 for mini-MIAS and 82.3 ± 4.78 for CBIS-DDSM database.