Mammogram Classification in Transform Domain

Abstract

Recently, breast cancer is major reason of cancer deaths among women. When cells in the breast tissue grow rapidly and gets divided without control breast cancer takes place which leads to formation of a mass or lump called as tumour. Here, the proposed method details comprehensive study and incorporation of image processing techniques to detect and classify tumours in terms of their accuracy. The proposed pre-processing technique removes all the unwanted labels present in an image to find region of interest (ROI). The various transformation methods such as Discrete Cosine Transform (DCT), Discrete Fourier Transform (DFT), and Radon Transform are applied to the ROI. Later, Gray-Level Co-Occurrence Matrix (GLCM) features are obtained. Lastly, the classification accuracy of detected abnormality is being found out using Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) classifiers. The recommended method is verified using Digital Database for Screening Mammography (DDSM) and Mammographic Image Analysis Society (MIAS) publicly available dataset. From the implementation, it has been inferred that out of three proposed methods a highest of 93.89% accuracy is achieved using combination of DFT and SVM classifier for DDSM database whereas for MIAS it returns 80% accuracy.