Medical Data Segmentation And Diagnosis Based On Intelligent Theory

Medical image segmentation and cancer diagnosis are essential parts of medical image analysis.Accurate segmentation of medical image provides very useful information for computer aided diagnosis and treatment of cancers as well as other diseases.Cancer diagnosis is the ultimate goal of medical image analysis.However,both automated medical image segmentation and cancer diagnosis face many challenges.According to the characteristics of medical image data,it can be divided into two categories: 1)images with obvious features;2)images with complex features.For those image images with obvious features,taking bladder wall segmentation as an example.Two potential problems are still not solved.1)due to the fact that noise and artifacts always exist inside the bladder.Those noise and artifacts are caused by the presence of urine and respiratory motion during imaging;2)the segmentation results may overflow at very weak boundaries.To solve the above problems,a shape prior constrained Particle Swarm Optimization(PSO)model is proposed.The proposed method can fully exploit the gradient information,shape prior information and inter-slice information to segment both the inner and outer boundaries of the bladder wall from MR images.The experimental results prove that our proposed model can significantly improve the performance of segmentation.For those image images with complex features,taking prostate segmentation as an example.In order to solve the following challenges during prostate segmentation:1)many slices only have small part of segmented tissues,specifically at the apex and base,which always led to those slices lack of clear boundary and make the automated segmentation fail;2)imaging artifacts distribute in the whole image randomly,which negatively influences the process of segmentation;3)segmented tissues have a wide variation in size and shape among different slices,which improves the difficulty of segmentation;4)the complex background and fuzzy boundary also make the segmentation process challenging;5)different from natural images dataset,the size of available med-ical image dataset is limited,which limits the learning ability of model.To overcome those challenges,we propose four kinds of Convolutional Neural Network(CNN)based models to address them.Model-1 is proposed to solve the problem that CNNs cannot fully extract prostate feature due to lacking enough training dataset.Model-2 can solve the problem that many slices lack boundary tissues specifically at the apex and base.Model-3 is designed to solve the problem that the long skip connections inside the network transfer the useless information.Model-4 employs transfer learning to overcome the problem of lacking sufficient training data.The above four methods continuously improve the accuracy of prostate segmentation.Cancer diagnosis also meets many challenges.Recently,disease diagnosis methods solely rely on the information from images and the relationship and similarity between subjects indicated by non-imaging features have been largely ignored.In addition,this kind of diagnosis methods also deviate from the process of actual clinical diagnosis.To tackle the challenges in the process of cancer diagnosis,we combine clinical data with imaging information and employ a novel Graph Convolutional Network(GCN)for feature extracting and diagnosis predication.