Study On Deep Learning Based Classification Of SPECT Images

Deep learning techniques,especially convolutional neural networks,have been widely used in reliable,accurate and feasible medical image classification because they can automatically learn image features in an optimal way.In order to reliably classify whole-body skeletal SPECT images and automatically diagnose SPECT images for a variety of diseases,this research has developed a set of multi-class classifiers for SPECT images based on deep learning.(1)Arthritis is a common and multiple physiological disease.Clinically,it is easy to make misjudgments between bone metastases,especially osteolytic metastases.In order to reliably identify arthritic lesions from SPECT images,with the help of the automatic feature extraction function of deep learning,this study constructed a SPECT image classifier for the automatic diagnosis of arthritis.First,normalize and expand the SPECT bone imaging data,appropriately expand the amount of data and convert it to the data format required by the model;then,build arthritis classifiers with different depths based on the standard VGG model;finally,use a Group real SPECT whole body bone imaging data to test the constructed classification model.Experimental results show that the classifier constructed in this chapter can effectively detect joint disease.(2)Aiming at the problem of whole body SPECT image classification,this study uses the well-known deep convolutional neural network.First,the original SPECT image is preprocessed through operations such as mirroring,translation,and rotation to enhance the original data set;then,on the basis of well-known deep learning models such as VGGNet,Res Net,and Dense Net,by fine-tuning their parameters and structure,Or customize a new deep network based on the structure of these models,and develop several multi-class classifiers;finally,through the experimental evaluation of a large number of real body bone SEPCT images,it shows that our classifier is effective for the classification of bone SEPCT images.The developed deep classifier can automatically classify the whole body bone SPECT image into the disease category of interest.(3)Multi-disease and multi-target classification of SPECT images based on deep convolutional neural networks.First,use traditional machine learning methods and generative adversarial networks to expand the data set separately;then,based on the research foundation of the previous two chapters,further construct a CNN classifier for SPECT images;finally,use a set of real bone SPECT images to evaluate the developed Deep classifier performance.The experimental results show that in the multi-disease multi-target classification experiment,the data extended by the traditional machine learning method has the best classification effect on the disease.