Research And Application Of CT Image Super Resolution Reconstruction Based On Generative Adversarial Network

Lung cancer is one of the malignant tumors with the highest mortality in the world.The high misdiagnosis rate and high missed diagnosis rate of lung cancer diagnosis directly lead to high mortality.Efficient and accurate diagnosis is the key to reduce the mortality of lung cancer.In the modern medical system,medical imaging technology plays an important role.Medical image’s clarity directly affects the accuracy of doctors’ diagnosis and Computer-aided diagnosis technology(CAD),which puts forward higher requirements on the quality of medical images.Super-resolution reconstruction can improve image quality and realize image enhancement.Therefore,this dissertation studies the super-resolution reconstruction of CT images based on generative adversarial network,and applies it to lung cancer subtype automatic classification task,the main work is as follows:(1)This paper proposes a super resolution reconstruction algorithm of CT images via a multi-image back projection-based generative adversarial network.In order to make full use of the sequence of CT image,Multi-path iterative projection units are introduced into the generator network.In addition,we construct discriminator and improve adversarial loss function under the guidance of WGAN-GP.The weighted sum of perceptual loss,content loss and WGANGP’s adversarial loss is used as the loss function of this model.The experimental results show that MBP-GAN can reconstruct medical images with higher subjective quality compared with other image reconstruction algorithms.(2)This paper proposes a classification model based on residual neural network for distinguishing pathological types of lung cancer.We improve Res Net-34 for lung cancer pathological types classification task and explored a medical-to-medical transfer learning strategy in the model.Specifically,the model is pre-trained on luna16,and then fine-tuned on our intellectual property lung cancer dataset collected in Shandong Provincial Hospital.The experiment results show that the proposed classification model has excellent classification ability in the case of insufficient data.(3)This paper applies the super-resolution reconstruction algorithm to the lung cancer pathological type classification model as data pretreatment step.First,the lung cancer dataset is reconstructed by the proposed super-resolution reconstruction algorithm for improving the image quality.Then,the reconstructed images are used as the input data of the lung cancer pathological type classification model.The empirical study’s result shows that adding MBPGAN to the lung cancer subtype classification algorithm improves the performance of the lung cancer subtype classification task,and ensures the reconstruction quality of medical images and the accuracy of subsequent analysis indicators.