Prostate cancer grading and classification by combining deep features and stochastic tissue features of pathological prostate images

Prostate cancer is one of the most important diseases of men whose growth can be disrupted by early diagnosis of it. In order to determine the grade of prostate cancer, the biopsy is used and structure of tissue is examined under microscopes. According to the new grading system, the prostate tissues are grading to five categories between 1 to 5 in which the high grades show the worst condition. Since human grading is time consuming, automatic grading systems have been used during recent years. Although some efficient algorithms have been introduced for image classification, the semantic gap between low-level features and human visual concept is still an important reason not to achieve high precision. In this paper, a new method for prostate cancer grading is presented which uses a combination of deep features extracted by convolutional neural network (CNN), and stochastic tissue features extracted using multi-level gray level co-occurrence matrices (ML-GLCM). Therefore, high-level features are achieved by using CNN and the grading precision is increased by combining the high-level features with stochastic tissue features. In order to evaluate the proposed method, it is experimented on the pathology prostate image database which is generated by international society of urological pathology (ISUP). Experimental results demonstrate that the proposed method achieved more accuracy than state-of-the-art methods on prostate cancer grading.