基于K空间采样的磁共振图像重建算法

ABSTRATWith the continuous development of medical imaging technology,MRI is more and morewidely used in clinical practice.However,because MR imaging is susceptible to local magneticfields when acquiring images,causing signal attenuation and distortion,thus affecting imagequality and diagnostic accuracy.Therefore,the application of MR image reconstructiontechnology in clinical practice is very important.Traditional reconstruction methods requiremultiple sampling and complex processing of data interpolation,filtering and normalization,andthe processing process is complex and easy to produce errors.During MRI data acquisition,k-space sampling is an important technical tool for efficient signal acquisition.However,someincomplete sampling situation often occurs in the k-space sampling process,resulting in the lossor corruption of the original data.Therefore,image reconstruction is a very important part inMRI data processing.To improve image quality and diagnostic accuracy,an efficient andaccurate image reconstruction method needs to be developed to overcome the shortcomings ofconventional methods in processing.Therefore,this paper designs a deep learning and MRI image processing technology,andadopts cascade dense network and multinomial data consistency constraint technology to solvethe problems of traditional methods.First,we used dense networks to reconstruct MRI images toimprove accuracy and robustness.Then,we further adopt the cascade structure and use multipledata consistency constraint technique to reduce image noise and improve image quality.Specifically,we use total variation regularization to reduce the variance of the data,making iteasier for the network to learn the rules in the data and improve the network performance,thusimproving the image reconstruction effect.In our experiments,we used several common MRIdatasets as test data,compared with conventional MR image reconstruction methods.The resultsshow that the present algorithm performs better in accuracy and robustness compared toconventional methods.Moreover,we also successfully applied this algorithm in some MRIimage reconstruction tasks in practice and achieved good results.In summary,this paper presentsan algorithm based on deep learning and MRI image,processing techniques,which has superiorimage reconstruction performance and robustness.The results of this study have importantsignificance for doctors'diagnosis and treatment.