Pancreatic CT Medical Image Online Segmentation System Based On Deep Learning

In recent years,computer-assisted therapy has become a hot topic.The use of deep learning methods in medical image processing has become a new research hotspot,which has promoted the development of the field of medical image analysis.Many traditional methods used in image processing have been replaced by deep learning segmentation networks.The main reason is that using deep learning for medical image segmentation usually produces higher accuracy.This article describes the existing segmentation technologies for medical images,including traditional methods and deep learning methods,and proposes improved methods for the deficiencies of the existing methods and deploys the improved algorithm model to the mobile client to complete the construction of an automatic pancreas segmentation system.Different patients have different pancreas shapes and blurred boundaries,so reliable automatic pancreas segmentation is an important and arduous task.In addition,there is currently no system dedicated to assisting doctors in pancreatic CT image segmentation in major domestic hospitals.In view of the above situation,the research work of this paper is as follows:(1)A two-stage cascade network is proposed for challenging pancreatic segmentation.First,the original CT image is roughly segmented through the first-level network,and then the result of the coarse segmentation is cropped and sent to the second-level network for further training to obtain the fine segmentation result.After cropping the CT image,irrelevant background interference is removed,and the input size of the second-level network is reduced,so that the segmentation accuracy can be well improved.(2)The saliency conversion function and attention mechanism are added to the cascade network.In the specific segmentation of the two-stage cascaded network,the coarse segmentation and fine segmentation stages are independent of each other,and the parameters are not shared.The saliency conversion function is added between the two stages to make the process of coarse segmentation to fine segmentation iterative and update the saliency conversion Module parameters participate in training together.In the process of network feature extraction,the loss of spatial information and channel information will inevitably occur.The addition of the attention mechanism can better focus and obtain higher accuracy in the region of interest.(3)Designed and implemented a mobile phone APP that can automatically segment medical images.The system adopts the above-mentioned improved neural network training model deployment.Select the abdominal CT image stored in the mobile phone,and display the segmentation result after training the network model with parameters.In the later design,if the user feels that the image segmentation effect is not ideal,he can upload his manually segmented CT image to the cloud and click to retrain.After the server receives the instruction,it will add the newly added image and label to the existing data set to re Training to achieve the effect of continuous iteration and update.