Semi-Supervised Deep Learning Networks In Medical Image Processing

Ultrasound imaging,X-ray imaging,magnetic resonance imaging,etc.are important medical imaging technologies,based on which doctors can effectively screen,diagnose,treat and evaluate in clinical treatment.In order to assist doctors to better analyze patients’ conditions,the development of medical imaging data has continued to improve imaging quality,noise processing,imaging efficiency and other issues.Today,the processing of massive amounts of high-quality medical imaging data poses new challenges for accurate and efficient analysis,especially for early screening of tumors and other conditions.The development of new techniques in the field of machine learning represented by deep learning has formed a new research hotspot in the field of image recognition.In this dissertation,we take COVID-19 CT image dataset COVIDx CT-3A as an example,and use various deep learning algorithms for classification and recognition,with the following main research aspects:(1)Establishing the classification model of deep learning algorithm for COVID-19 CT image dataset,this dissertation compares the recognition effect of four deep learning algorithms,which are widely used at present,on the CT image dataset of novel coronavirus pneumonia based on the idea of fully supervised learning,and the experimental results show that the better recognition effect on the CT image dataset of novel coronavirus pneumonia is finally obtained through the parameter adjustment The fully-supervised network model was obtained.(2)Designing a deep learning algorithm based on semi-supervised learning to accurately classify the images of the CT image dataset of novel coronavirus pneumonia.The experimental results show that the semi-supervised network model can perform accurate classification under the condition that the percentage of labeled data in the CT image dataset is relatively low.The increasing arithmetic power of computer technology provides a means for deep learning algorithms to provide both efficiency and accuracy in medical image classification and recognition,which has great potential for application in aiding the diagnostic work of CT images of multiple diseases.(3)Visualization of the region of interest of the deep learning algorithm model,in this dissertation,the algorithm achieves better recognition results,while investigating why the algorithm arrives at this result,the experiment involves how the data distribution and features of the CT image dataset of novel coronavirus pneumonia affect the classification results and how the recognition classification results of different classification models are comparatively interpreted and evaluated,the experimental results show the connection between the region of interest of the network and the recognition The experimental results show the connection between the network region of interest and the recognition results.