Automatic breast cancer analysis using Bandelet transform

This paper presents a mammographic analysis using Bandelet transform for breast cancer diagnosis. Many multiresolution approaches are available for breast cancer detection from the mammograms. In this work, Bandelets are explored for the analysis of mammograms. In Bandelet transform, the image is decomposed along multiscale vectors that are elongated in the direction of a geometric flow. This geometric flow indicates directions in which the image gray levels have regular variations. Feature vector is formulated by computing the mean, standard deviation, skewness and kurtosis of the Bandelet coefficients. Support Vector Machine (SVM) is chosen as the classifier. Different classifiers are explored. Extensive simulations are carried on Mini-Mias database. The Mammograms were classified on 80-20 fold classification for Abnormality analysis i.e classifying as normal or cancer. The Mammograms are classified based on type of breast tissues i.e Fatty, Fatty Glandular and Dense and are also classified based on type of cancer i.e Microcalcification, Spiculated, Asymmetry, Architectural Distortion, Circumcised and Miscellaneous. Performance of the proposed algorithm is assessed using classification accuracy, confusion matrix, precision and recall. Compared to Wavelets and Curvelets, higher classification accuracy is obtained using Bandelets. Further, results are compared with the existing methods and the superiority of the proposed method in terms of classification accuracy and other metrics is demonstrated and justified.