Segmentation And Recognition Of Thyroid Associate Ophthalmopathy Based On Deep Learning

With the rapid development of the electronic information technology,the technical of modern hospitals has also been significantly improved.The advent of high-tech has made corresponding medical operations automated,eliminating the need for manual and complicated operations.Thyroid Associate Ophthalmopathy is one of the common orbital diseases in adults.It is an autoimmune disease,and the exact pathogenesis is not clear.The distortion of the rectus eye muscle is one of the main causes of TAO.Segmentation and detection of the rectus eye muscle are of great significance for the initial screening and later diagnosis of TAO.This thesis uses deep learning technology to conduct automatic segmentation and detection of rectus eye images,which can reduce labor costs and improve diagnostic efficiency.The first research point of this thesis is to solve the problems of excessive convolutional neural network parameters and poor segmentation effect.The U-Net based on dense convolutional neural network Dense Net and selective convolution kernel mechanism SKNet is proposed to segment the CT image of rectus muscle.In the U-Net model,the encoder uses the pooling layer to reduce the spatial dimension of the input data,the decoder recovers the details of the target and the corresponding spatial dimension through the deconvolution layer and other network layers.There is a direct information connection from the encoder to the decoder to help the decoder better recover the target details.In the traditional convolutional neural network,the receptive field of each layer of artificial neurons is fixed,but in the field of neuroscience,the size of the receptive field of visual cortical neurons will be stimulated and adjusted to selectively convolve the kernel.The mechanism SKNet can make each neuron adaptively select the size of the receptive field according to the multi-scale information of the input features,and improve the segmentation accuracy.The second research point of this thesis is that in the task of eye CT image detection.In order to solve the problem of difficult recognition and small sample number in CT images of thyroid-related ophthalmopathy,the CT image recognition method based on the transfer learning is proposed for thyroid-related eye diseases.Transfer learning can be applied to the data in the original field very well,and the model can be used in other fields,which can save a lot of time and resources.Using transfer learning to construct a convolutional neural network to complete the final detection of the rectus eye image.In summary,this thesis studies the U-Net network based on Dense Net and SKNet to segment the eye CT image,then the segmented rectus CT image is put into the fine-tuning migration model Inception-V3 to train,to complete the finally detection of the TAO image.The effectiveness and practical application value of the above-mentioned models and methods are verified through experiments,it also has a good reference for other types of medical image classification.