Breast Cancer Diagnosis from Biopsy Images Using a Fully Automatic Method

The most reliable way to diagnose breast cancer in the current practice of medicine is through pathological examination of a biopsy which has a certain level of subjectivity. To reduce this subjectivity and have a mathematical model for diagnosing breast cancer tissues, a fully automatic method based on microscopic biopsy image is presented. The novel technique is based on a four-step procedure: the pathologic images are de-noised and enhanced based on k-nearest-neighbor (KNN) and histogram equalization method; morphology features are extracted using wavelet moment invariants; a rough set (RS) is applied to reduce features dimensions and select the best features; a multi-category proximal support vector machine (MPSVM) is designed to reliably differentiate normal, in situ and invasive breast cancer tissues. The experiments demonstrate that the proposed method is effective and useful for classifying breast tumors.