Uterine Fibroids And Carotid Plaque MRI Dichotomous And Multi-classified Auxiliary Analysis System

With the continuous improvement of computer computing performance and the massive increase of medical image data,neural network has been widely used in medical image segmentation.Compared with the manual segmentation by experienced doctors,the deep neural network can greatly shorten the segmentation time and improve the efficiency of doctor's reading medical images by automatically learning local features through multi-layer network.Therefore,combining the advantages of deep neural network and MRI images,this paper will study the segmentation techniques of dichotomy and multi-classification of MRI images based on deep neural network,and specifically select the MRI images of uterine fibroids and carotid artery plaques.The details are as follows:1.For dichotomous MRI T2W image segmentation of uterine fibroids,a kind of MRI T2W image segmentation network AR-Unet(Attention Res Net101-Unet)was designed.We combine the design ideas of U-Net,use Res Net101 main body for feature extraction,and add a designed attention module to improve the segmentation accuracy before the upsampling and downsampling feature maps of the same layer are spliced.In addition,the loss function is improved,and a loss function L_NDice suitable for AR-Unet network is designed.We randomly selected 947 MRI T2W images of uterine fibroids from 80 patients and put them on the network for training,so that the loss value of the network fell below 0.05,and a total of 123 fibroid MRI T2W images from 13patients were tested using the trained network,the segmentation results were compared and verified with the expert segmentation results,the average Dice coefficient,Io U value,sensitivity and specificity of all segmented images are respectively 0.9044,0.8443,88.55%and 94.56%.The performance is superior to Res Net101-Unet and Attention U-Net models.Finally,the network is packaged into the GUI interface named MRI image assisted segmentation and classification system.2.For multi-classification MRI image segmentation of carotid artery plaques,an improved U-Net MRI image segmentation network was designed.Four types of images,TOF,T1W,PDW and CTA,were used as network inputs.Four separate paths were used in the downsampling section to process different modal images separately,with super-dense connections in each path and across paths,and an improved Inception module was added within each path.The upsampling is carried out by means of transpose convolution.The normalization of network is based on FRN(Filter Response Normalization)and TLU(Thresholded Linear Unit)as the activation function.We randomly selected 24 patients with a total of 240 images to train the network,so as to achieve the optimal network performance.58 MRI images of carotid plaques in 6patients were tested,the segmentation results were compared and verified with the expert segmentation results,the average Dice coefficient,Io U value,sensitivity and specificity of all segmented images are respectively 0.8182,0.7123,78.13%and85.04%.The performance is better than U-Net network.Finally,the network is packaged into the GUI interface named MRI image assisted segmentation and classification system.