Research On The Recognition And Segmentation Algorithm Of Vascular Infarction In MRI Cerebral Images

Cerebral infarction,also known as cerebral infarction,refers to a large area of cerebral ischemia and hypoxic necrosis due to impaired blood circulation in the brain.In clinical diagnosis,when MRI is used to determine cerebral infarction,the diseased tissue has the characteristics of complex shape,variable size and blurred boundaries.Therefore.experienced clinical technicians are required for observation and manual annotation.and there are often problems with different diagnosis results due to different technicians.At present,deep learning has been widely used in medical image processing,using deep learning methods to identify and segment the location of lesions in cerebrovascular infarction,thus replacing artificial technicians.Accelerate the speed of clinical diagnosis and reduce the occurrence of misdiagnosis and missed diagnosis.Therefore,it is very necessary to study multi-scale target detection algorithm and adaptive precise segmentation algorithm.This article first clarifies the significance of theoretical research and clinical practice,and then introduces the research status at home and abroad.Secondly,focus on the relevant knowledge of this topic.Then it focuses on the algorithm proposed in this article,including mathematical reasoning and algorithm operation steps.Finally,based on the experimental results,make evaluation analysis.The main innovations of this paper are as follows:(1)Aiming at the problem of brain MRI image detection and recognition of cerebral infarction,a Cerebrum R-CNN framework is proposed,Extend the 2D R-CNN network to 3D medical image detection,add a variety of residual modules to the feature extraction network FE-Net and use the nearest neighbor interpolation method,the feature alignment module CerebrumRoIAlign block adds trilinear interpolation mode and classification unit Softmax3,region It is recommended to use the classification unit Softmax2 in the network CerebrumRPN to divide the brain tissue to which the output RoI region belongs-At the same time,the secondary alignment mechanism is used in the bounding box prediction network CerebrumBox to ensure the accuracy of the output candidate box.Compared with traditional recognition methods,this paper proposes that the Cerebrum R-CNN model has an accuracy rate of 86%,a recall rate of 79%,and an IoU of 87.1%for the detection of this experimental data set(2)Aiming at the problem of segmentation of cerebral infarction in MRl images of brain.the adaptive segmentation CerebrumFCM algorithm is improved and proposed.Based on the classic FCM(Fuzzy C-Means)algorithm,the local local constraint SF FCM algorithm is proposed.Quantum particle swarm and space conversion ideas are further introduced.At the same time,the local grayscale statistical method is added,and then the Cerebrum FCM algorithm is proposed.The adaptive segmentation CerebrumFCM algorithm proposed in this paper has a Dice coefficient of 84.12%on this experimental data set,and its performance exceeds the existing FCM and its improved algorithms.