Research On Ultrasonic Image Recognition Method Of Prostate Cancer Based On Deep Learning

Early diagnosis of prostate cancer can greatly improve the survival rate of patients,and the complexity of the diagnostic procedure affects the accuracy of early diagnosis of prostate cancer.In the diagnosis technology of prostate cancer,transrectal ultrasound has the advantages of low cost,non-invasiveness and simple operation,and can be widely used in early detection.However,the specificity of transrectal ultrasound for prostate cancer is low,and it is not easily distinguished from echoes of other prostate diseases.The accuracy of diagnosis and guidance of puncture depends on the doctor’s experience and skills,and has a certain degree of subjectivity.At present,deep learning technology has made great progress in the field of image recognition.This article applies deep learning technology to the method of prostate cancer ultrasound image recognition.The specific research content is as follows:Research on the segmentation method of prostate ultrasound image based on DeepLabv3+.Aiming at the segmentation of the target region of transrectal ultrasound image of the prostate,this paper studies the prostate region segmentation method of the prostate ultrasound image based on the convolutional neural network model.In this paper,DeepLabv3+ is used as the image segmentation deep learning algorithm,Image Net pre-training weights are used as the initial weights of the segmentation model,and a relatively small number of ultrasound image data sets are used to fine-tune the model weights,so as to let DeepLabv3+ learn the experience and experience in other image tasks.Knowledge can be transferred to the ultrasound image segmentation task to solve the problem of too few training data sets and improve the accuracy of model segmentation.Research on ultrasound image classification technology of prostate cancer based on ResNet.Aiming at the problem that the existing convolutional neural network image classification algorithm has low accuracy in identifying prostate cancer based on transrectal ultrasound images,this article adds an image segmentation step before image classification.In this paper,the single-channel Mask image with segmentation information and the three-channel transrectal ultrasound prostate image are combined and superimposed into a four-channel image.However,mainstream image classification algorithms cannot support the input of four-channel images.This paper uses the ResNet network model as the classification Deep learning algorithm,the input layer of the ResNet classification model is improved,and the image input channel is changed to four channels,so that the classification model can support transrectal ultrasound prostate images with segmentation information.Experimental results show that the above improvements greatly improve the accuracy of prostate cancer ultrasound image classification.The ResNet classification model is improved on the basis of the VGG classification model.The comparison experiment with the VGG19 classification model also proves that the ResNet classification model of this paper has a high accuracy rate.