Research On 3D-MRI Image Super-resolution Reconstruction Based On Deep Convolution Generative Adversarial Network

Three-dimensional magnetic resonance imaging(3D-MRI)is a common medical imaging technology that can be applied to the diagnosis of diseases in various systems throughout the body and has important application value in the medical field.At present,the main disadvantage of the three-dimensional magnetic resonance imaging technology is that it takes a long time for scanning and imaging,and inspection is time-consuming.Therefore,for patients,it is not only difficult to maintain a static state in a claustrophobic space for a long time during scanning,but also easily leads to queues of patients.Problems such as too long appointment time.In order to reduce the time required for scanning and imaging,the usual clinical practice is to increase the thickness of the scan,but at the same time it will reduce the number of scans,which will then reduce the resolution of the other two dimensions of the 3D-MRI image,which will ultimately affect the doctor’s diagnostic analysis and analysis.Post-processing of images.Therefore,improving the resolution of images is a topic that 3D-MRI has to solve all the time.Super-resolution(SR)reconstruction research can effectively solve the above problems,that is,reconstructing high-resolution images that meet the requirements based on low-resolution images.At present,the research on SR reconstruction of 3D-MRI images has certain challenges.Often the quality of the reconstructed image cannot reach the expected ideal effect,and there is still much room for improvement in the quality of the reconstructed image.At the same time,some SR reconstruction methods also have problems such as complex network models,huge parameters,high hardware configuration requirements,and image quality evaluation methods that do not conform to human visual perception mechanisms.Therefore,aiming at the current problems in the field of 3D-MRI image SR reconstruction,this study adopts Enhanced Super-Resolution Generative Adversarial Networks(ESRGAN),combined with the cross-layer self-similarity features of 3D-MRI images,and builds Deep Convolutional Generative Adversarial Networks(DCGAN)3D-MRI image SR reconstruction model,SR reconstruction of 3D-MRI images.The model first slices the 3D-MRI image along different planes(cross-section,coronal plane,and sagittal plane)into two-dimensional magnetic resonance images(2D-MRI)through image slicing operation;The image SR reconstructs a high-resolution 2D-MRI image;the reconstructed high-resolution 2D-MRI image is restored to a 3D-MRI image according to the original index sequence;finally,the high-resolution 3D-MRI image will be obtained from different levels Perform image fusion to obtain the final high-resolution3D-MRI image.Through this model method,the SR reconstruction task is reduced to two dimensions,which can not only increase the image training data set,but also reduce the amount of model parameters,thereby reducing memory requirements and speeding up model training.The experiment uses the publicly available brain clinical real data Kirby21 as the training data set to train the 3D-MRI image SR reconstruction model proposed above;the Bra TS data set containing gliomas and the simulated data Brainweb data set are used as the test data set.The trained reconstruction model was tested;the quality of the 3D-MRI images generated by the experiment was evaluated,and four commonly used objective evaluation indicators in the field of image SR reconstruction(peak signal-to-noise ratio,structural similarity,root mean square error,and perception index)carried out image quality evaluation;the experiment also carried out a visual display of the reconstruction results;finally compared the reconstruction results of five existing methods in the field of image SR reconstruction research,and did a quantitative and qualitative analysis of the experimental results.Through comparative analysis with the results of five common SR reconstruction methods,the model constructed in this study can achieve better visual perception effects,indicating that the model has certain advancement and superiority.The experimental results show that the model can achieve better performance in the evaluation of perception index,which is basically in line with the human visual perception mechanism.In addition,this model not only has an excellent SR reconstruction effect for3D-MRI images of the same modality,but also shows relatively excellent performance for SR reconstruction of 3D-MRI images of different modality.Through the application of this model,the resolution of the remaining two-dimensional images after 3D-MRI image reconstruction can be effectively improved,which is conducive to the diagnosis and analysis of diseases and post-processing of images by doctors,and is conducive to reducing patient waiting in 3D-MRI image examinations time.Therefore,the SR reconstruction model has good application potential in 3D-MRI image inspection.