Research On The Technology About Ultrasound Image-aided Diagnosis Of Prostate Cancer Based On Deep Learning

Prostate cancer is one of the most common malignant tumors,the number of cancer is increasing year by year,but the majority of patients do not have effective treatment.Computerized diagnosis of prostate ultrasonography is helpful for the detection and treatment of prostate cancer.However,there are some problems in prostate ultrasound image,such as unclear boundary,small target area and low resolution,large difference in shape and size of lesion area.In order to overcome these difficulties,this paper proposes a deep learning based ultrasound image assisted diagnosis technology for prostate cancer,which can significantly improve the diagnostic accuracy of prostate cancer.In view of the above difficulties,this paper focuses on the deep learning-based ultrasound image detection and diagnosis model for prostate cancer and its efficient training methods.The main work is as follows:Studying on segmentation technology of prostate ultrasound image based on S-Mask R-CNN.Segmentation is a key step in the data enhancement of prostate ultrasound images,which can reduce the image information to the key part and eliminate the interference of the too much irrelevant information.The improved S-Mask R-CNN was used to achieve accurate segmentation of prostate ultrasound images.The region of interest(ROI)was used to achieve pixel level feature point localization.The binary mask corresponding to the prostate region from the background.Then,several groups of prostate ultrasound segmentation image data sets were constructed to be used in the Inception-V3 network for image classification and detection by shielding the background information.The experimental results show that the algorithm proposed in this paper has a good segmentation effect under the premise of reasonable time consuming.Classification of prostate ultrasound images based on Inception-V3.In the Inception-V3 network,with a new classification module consisting of forward propagation and back propagation to replace the original network softmax the convolutional layer below pool-3layer to the input layer through the transfer learning strategy,and then calculate the loss value between the classification value and the label to identify the prostate ultrasound lesions.The experimental results show that compared with the classical classification methods and manual diagnosis,the proposed method is more accurate in classifying the negative and positive of prostate ultrasound images.Studying on marking of ultrasound images of prostate cancer based on improved S-Mask R-CNN+Xception.Gleason grading of prostate cancer provides important reference value for clinical evaluation of treatment plan.However,there are few studies on pathological grading of ultrasound images of prostate cancer.How to accurately describe its aggressiveness is a hot issue.Therefore,this paper proposes a Gleason grading network model for prostate cancer based on deep learning.The model consists of three modules: lesion area labeling,image enhancement and pathological grading.To be specific,the improved S-Mask R-CNN network model was used to mark the focal areas of prostate cancer,and then image enhancement was performed through morphological image processing and PCNN algorithm.Finally,Xception was used for pathological grading of prostate ultrasound images.The experimental results show that the prediction accuracy of Gleason grading model in this paper is 0.28 higher than that of other classical grading models and manual diagnosis by doctors.According to the above three steps,this paper built an intelligent assisted diagnosis system for prostate cancer ultrasound imaging to carry out experiments.The experimental results show that: In this paper,the method can accurately detect prostate cancer at the same time in the pixel level of ultrasonic image information,compared with the diagnosis and other medical professional manual detection method,the method has higher accuracy,this method will be used as a solid foundation,and simplifies the future computer aided diagnosis of prostate cancer ultrasound image research,this work will promote the development of prostate cancer ultrasound image diagnosis technology.