A combined approach of multiscale texture analysis and interest point/corner detectors for microcalcifications diagnosis

Screening programs use mammography as primary diagnostic tool for detecting breast cancer at an early stage. The diagnosis of some lesions, such as microcalcifications, is still difficult today for radiologists. In this paper, we proposed an automatic model for characterizing and discriminating tissue in normal/abnormal and benign/malign in digital mammograms, as support tool for the radiologists. We trained a Random Forest classifier on some textural features extracted on a multiscale image decomposition based on the Haar wavelet transform combined with the interest points and corners detected by using Speeded Up Robust Feature (SURF) and Minimum Eigenvalue Algorithm (MinEigenAlg), respectively. We tested the proposed model on 192 ROIs extracted from 176 digital mammograms of a public database. The model proposed was high performing in the prediction of the normal/abnormal and benign/malignant ROIs, with a median AUC value of 98.46% and 94.19%, respectively. The experimental result was comparable with related work performance.