Research On The Algorithm Of Processing And Classification In Near Infrared Breast Images

Breast cancer is one of the most common malignancies in the world,and there are no effective cure measures so far.As the disease incidence of breast cancer increased every year,early detection and diagnosis is significant to reduce the mortality of breast cancer patients.Near infrared spectroscopy is one of the most important methods for the early detection of breast diseases.However,the accuracy of diagnosing breast cancer in images by doctor's eyes is poor,even though the doctor has abundant professional experience.Thus,to process near infrared breast images and classify the cancerous parts automatically,which is the focus of this thesis,to assist doctors is of great significant.The main research work and achievements are as follows:(1)Weighted-guided filtering based on Canny operator to pre-process the image proposed in this thesis,compared with the weighted-guided filtering based on variance,has excellent performance.When dealing with the image,different window radius and regularization factor are used to reduced the noise and strengthen the edge texture information.The method provides by this thesis obtains an outstanding result,and verifies the advantages of this algorithm.(2)Then,the fuzzy C-means clustering algorithm is used to segment the image after preprocessing.And in this thesis,we analyze the initialized parameter of the network,so that the network could separate the cancerous parts from the normal breast organization automatically.The segmentation result is obvious,which can assist the doctor effectively.(3)This thesis provides a VGG(Visual Geometry Group)convolution neural network with 11 layers.The network is trained on near infrared breast images,tested different image blocks and determined the best size of image blocks.The classification accuracy is good.Compared with the algorithm of FCM,VGG does not need to set various parameters previously,meanwhile,it can find the best parameter by iteration continuously,which makes the network more robust and superior.The experimental result shows that the VGG convolution neural network not only classify the near infrared breast images,but also has high accuracy which is up to 81%.