Automated mitosis detection based on combination of effective textural and morphological features from breast cancer histology slide images

Number of mitotic cells in histopathology images of breast cancer is graded as one of three important factors for this cancer. In this paper, an automatic system is presented to assist pathologists in the diagnosis and mitotic cell counting in less time to overcome two major challenges in this context. The first challenge is the large variety in the structure of mitotic cells and the other is the large number of candidates for mitotic cells. In the proposed method, statistical Gamma-Gaussian Mixture Model (GGMM) has been employed in extraction of primary candidates, for estimating the Probability Density Function (PDF) of mitosis and non-mitosis cells. The effective features based on the texture and shape using Gabor filter, co-occurrence Matrix (GLCM), Complete LBP and morphometric and shape-based features, extracted from each candidate. Finally, by using of Support Vector Machine (SVM) classifier with different kernel functions of RBF, Linear and Quadratic and decision tree classifier with 50 trees, mitotic cells has been detected. The evaluations are applied over histology datasets A and H offered by the Mitos-ICPR2012 contest sponsors. After calculating the evaluation criteria, the results of the proposed method has achieved F-measure 92.3077% for scanner images Aperio XT and F-measure 89.4009% for scanner images Hamamatsu, that reflect the performance of the proposed method.