Research On Enhancement And Detection Of Hemangioma And Hepatocellular Carcinoma Based On Generative Adversarial Network

Contrast-enhanced magnetic resonance imaging(CEMRI)is crucial for the diagnosis of patients with liver tumors,especially for the detection of hemangioma and Hepatocellular carcinoma.However,it suffers from high-risk,time-consuming,and expensive in current clinical diagnosis due to the gadolinium-based contrast agent(CA)injection.If the CEMRI can be synthesized without CA injection,there is no doubt that it will greatly optimize the diagnosis.In this paper,we propose a Multi-view and radiomics-guided generative adversarial network for enhancement of hemangiomas and hepatocellular carcinoma(HCC).And we propose a tripartite adversarial generative adversarial network to synthesize contrast-enhanced magnetic resonance image for facilitating the detection of liver tumors.In the work of multi-view and radiomics-guided generative adversarial network(Mv Rg-GAN)for enhancing the specificity of hemangioma and hepatocellular carcinoma without contrast agents.The main research content and contribution as follows:i.For the first time,Mv Rg-GAN successfully synthesizes liver CE-MRI by integrating multi-modality NC-MRI information without CA injection.It provides an AI-CA as a safe,time-saving,and inexpensive tool for enhancing the specificity of hemangioma and HCC to facilitate clinical diagnosis.ii.The multi-view and attention-aware generator fully-adaptively integrates the multi-modality information of T1 FS pre-contrast MRI and DWI by using the innovative fusion valve,which enhances the specificity of hemangioma and HCC.And newly design Res-GAM makes Mv Rg-GAN context-aware to enhance the feature representation and spatial continuity.Besides,the residual connection of Res-GAM also improves the convergence of the training loss for the MvRg-GAN to be easily trained.iii.A novel radiomics-guided discriminator improves its own ability to distinguish the RCE-MRI and SCE-MRI via adding prior knowledge of RF.And it promotes the generator to synthesize high-quality and accurate liver CE-MRI via adversarial-strategy.iv.A Sobel-based edge detector is incorporated into Mv Rg-GAN to obtain edge information,which is used for enhancing the boundary synthesis and the continuity of SCE-MRI.In the work of synthesizing liver contrast-enhanced MRI to improve tumor detection via tripartite-GAN.The main research content and contribution as follows:i.For the first time,synthesizing CEMRI without CA injection for liver tumor detection is achieved,which provides a safe,time-saving,and inexpensive clinical tool to synthesize CEMRI without CA injection.ii.The newly proposed Tripartite-GAN successfully combined the regular two-participant GAN and the detector via back-propagation for the first time,which achieves that CEMRI synthesis and tumor detection promote each other in an end-to-end framework.iii.The newly designed attention-aware generator is powerful in feature ex-traction with the help of hybrid convolution,residual learning,and DAM.Specifically,the hybrid convolution enlarges the receptive field efficiently,the residual learning benefits the convergence to facilitate the training of the generator,and the DAM enhances feature representation learning of tumor specificity and context learning of multi-class liver MRI.iv.Attention maps from the generator newly added into the detector in the manner of residual connection improve VGG-16 based convolution operation to extract tumor information better,which improves the performance of tumor detection.The experimental results prove that our method can provide a time-saving,non-invasive and zero-cost technology for clinical diagnosis of hepatic hemangiomas and hepatocellular carcinoma.Moreover,we propose a y-shaped Unified Network(y-Net),which is the first to achieve simultaneous segmentation and detection of HCC using multi-modality NCMRI only.The y-Net mainly goes through three innovative parts: 1)the y-Net enables simultaneous HCC segmentation and detection in an end-to-end framework;2)the newly designed Information Filter makes framework multi-view aware via fully-adaptively integrate and select the complementary information among multi-modality NCMRI;3)the innovative Task-related Channel achieves mutual promotion between HCC segmentation and detection via deploying the complementary feature and back-propagation.The main research content and contribution as follows:i.For the first time,y-Net provided a time-saving,safe,and inexpensive tool,which achieves simultaneous segmentation and detection of HCC via using multi-modality NCMRI only.ii.Newly designed Information Filter provided a novel solution to integrate and select the complementary information among multi-modality image for multi-task.iii.Innovative Task-related Channel enables the mutual promotion between multi-task via deploying the complementary feature and back-propagation.