Gleason Grading of Prostate Cancer Based on Improved AlexNet

Prostate cancer is a common malignant tumor in male genitourinary system, its morbidity is increasing in recent years. Puncture pathological examination with Gleason scoring is the ultimate means of diagnosing prostate cancer. Early detection of prostate cancer is obviously very important for the treatment and prognosis of the cancer. However, the pathological image of prostate cancer has a complicated texture structure, especially the difference between Gleason Grade 3 and Gleason Grade 4. Therefore, pathological images with a Gleason score of 7 are difficult to distinguish between "3+4" and "4+3". The misjudgment of "3+4" and "4+3" impact on quality of life in patients with prostate cancer after operation can be profound. In order to improve the classification accuracy of histopathological images of prostate cancer especially for detecting "3+4" and "4+3", this paper proposed an image classification model based on improved AlexNet. On the basis of ALexNet, Res1_block and Res2_block structures are added to extract the features of pathological images. Experimental results show that our approach can automatically classify prostate cancer pathological images, and the test accuracy can reach 78.4%.