Intelligent Analysis Of Cardiac Vascular Image Morphology And Function

In the medical field,the analysis and processing of medical images plays an important supporting role in the diagnosis of diseases by doctors.For many years,scientists have been paying attention to processing techniques such as pattern recognition,classification and segmentation in medical image analysis.Machine learning techniques enable researchers to develop and utilize complex models to classify or predict various abnormalities or diseases or to identify and segment medical lesions.In recent years.the rapid development of deep learning has made the auxiliary diagnosis of medical images to the peak.In this paper,the morphology and function of cardiovascular images in medical images are analyzed based on the method of deep learning,which is of great significance for further analysis of diseases and assistant diagnosis.One of the research challenges encountered in this paper is that due to the complexity and confidentiality of medical images,there are currently few data sets for disclosure,and because the label of medical images requires professional evaluation.there are fewer data sets with labels.The other is because medical images are generally grayscale images,and there are few features.In traditional machine learning,features based on grayscale textures are used,which makes the analysis very limited.According to this,the three main contributions of this paper are as follows::(1)Some series of data sets of intravascular ultrasound images have been constructed.The data in this paper is clinical data collected from hospitals and is of great clinical significance.The first is the data set of human normal blood vessels and bifurcated blood vessels.This data set includes 2288 images,including 1144 normal blood vessel images and 1144 images of bifurcated blood vessels.The bifurcation vessels are labeled under the guidance of a professional doctor.Then the lumen and media-adventitia segmentation data sets of bifurcated vessels and normal vessels were constructed respectively,including 1144 bifurcation blood vessel data and 5216 normal blood vessel data.The labeling of the lumen and media-adventitia in the data set is marked by professional medical students and has certain medical credibility.Finally,the functional analysis data set of cardiac vascular images was constructed based on hospital biological experimental data.Twenty-three vascular pathological sections of the rabbit atherosclerotic model and corresponding 3535 intravascular ultrasound images were acquired.(2)An intravascular ultrasound image processing system based on deep learning is proposed.In this system,the deep learning method is applied to the classification of IVUS images and the segmentation of lumen and media-adventitia in medical images.Firstly,based on the four classical classification networks AlexNet,VGG,ResNet and DenseNet.the bifurcation vessels and normal blood vessels are classified,and evaluate the classification performance of the four models to get the best classification model.The classification results,that is,the bifurcation vessels and the normal blood vessels,use the four segmentation networks FCN,DeepLab,and GAN to separate the lumen and media-adventitia,evaluate the segmentation performance of the four segmentation networks.Save the network with the best segmentation effect and output the segmentation result.Finally,the obtained endometrial segmentation results are reconstructed in three dimensions for further analysis to assist the physician in disease analysis and diagnosis.(3)Based on deep learning,a new method for determining the boundary between vulnerable plaque and stable plaque vulnerability index is proposed.The medical image is further analyzed.First,the IVUS image of the rabbit was collected,and the collected blood vessels were made into pathological sections to calculate the vulnerability index.Once a vulnerability index was determined,and the label was made based on the vulnerability index.Above this vulnerability index,it is marked as a ruptured plaque.that is,an unstable plaque,and below this vulnerability index,it is considered to be a stable plaque.Based on the labeled IVUS images,the convolution neural network is used to classify them,and then the vulnerability index points are replaced continuously to obtain the corresponding classification accuracy.According to the fitting relationship between the vulnerability index and the accuracy rate,the vulnerability index with the highest classification accuracy is found.We think this is the best vulnerability index classification point.According to the experimental results,the critical point of the optimal vulnerability index is 1.022.However,when trying to expand the range of vulnerability index,it is found that the fitting curve is no longer applicable.Therefore,the fitting function is replaced by Fourier transform,and the fitting curve has a certain periodicity.After verifying the verification set,it is found that the accuracy of the verification set reaches its maximum when the vulnerability index is 2.398.When the data source is changed from IVUS image to patch component data as classification data,it can be found that the rule still exists.However,the best vulnerability index found at this time decreased slightly to 2.198.Finally,the classification points of vulnerability index are verified based on the verification set,and the highest accuracy is 76.5%.This shows that this method has certain effect in finding the best classification vulnerability index points.