Data Augmentation Based Low-dose CT Reconstruction Algorithm With Priori MRI Images

Computed tomography(CT)is one of the most commonly used medical imaging methods.CT utilizes penetration of X-ray through objects,and the excessive radiation dose can cause harm to the human body,so it is necessary to reduce the X-ray dose during CT scanning.The dose can be reduced by reducing the tube voltage,tube current,or the number of projection angles.However,this will result in poor imaging quality.Hence how to improve the imaging quality in low-dose CT is a major challenge for medical industry.In order to improve the quality of low-dose CT imaging,magnetic resonance imaging(MRI)at the same location can be used as a structural priori MRI image(SPMI),supplemented by registered CT-MRI connection dataset.The existing priori MRI based CT reconstruction algorithms assume that CT and MRI are integrated into one machine and simultaneously scanned.In this way,CT image and SPMI can be well registered,otherwise the reconstruction quality of existing methods will be seriously reduced,but no such machine has been successfully introduced.In order to be better applied to the clinic,this paper considers that CT image and SPMI are from different machines.In this case,inter-modality well-registration between SPMI and low-dose CT image is a complex and time-consuming task,which can not guarantee the accuracy of registration either.Accordingly,this paper focuses on how to improve the image quality of low-dose CT with roughly registered SPMI and CT image.There may be rigid and non-rigid structural differences between roughly registered CT and MRI images.Rigid differences mainly refer to position and angle,which can be simulated by translation and rotation,respectively.Non-rigid differences are more complex,thus in order to simplify the problem,this paper mainly considers the influence of scale.For convenience,position,angle and scale are abbreviated as PAS when necessary.For the differences of PAS,augmented dataset based low-dose CT reconstruction algorithm with priori MRI images is proposed by the idea of data augmentation in this paper.This method introduces the relative position,angle and scale information between CT and MRI images by augmenting the connection dataset.The differences in PAS between SPMI and low dose CT image can be global or local.The global PAS difference means that SPMI can be well registered with CT image through translation,rotation or scaling;The local PAS difference refers to the incorrespondance in position,angle and scale between SPMI and CT image in a certain local area,while non-rigid deformation exists in other areas.In this paper,experiments are designed on two sets of datasets to verify our method to improve the imaging quality of low-dose CT in the case of global or local PAS differences,and in the case of well-registration at the same time.The experimental results show that using priori MRI image can improve the quality of low-dose CT imaging.And if there are global PAS differences between SPMI and CT image,the proposed method can effectively improve the quality of CT image,and has strong robustness.With the increase of global PAS difference,the imaging quality will not be significantly reduced,and the root mean square error(RMSE)can always be kept in a relatively small range.According to experiments,RMSE can reach 0.0251 when SPMI is well registered with CT image.And RMSE can be maintained between [0.0251,0.0275],[0.0251,0.0259] and [0.0251,0.0269],respectively,if the left or right position deviation is within 4 pixels,the angle deviation is within 6° or the scaling factor is between [0.96,1.04].On the other hand,if there are local PAS differences and other regions have non-rigid deformation,the proposed method can also effectively improve the quality of low-dose CT image.Specifically,for regions with local PAS differences,the improvement of CT image quality is very significant,while the image quality of other regions can also be improved.Although only three basic structural differences are considered in this paper,it lays the foundation for the study of other non-rigid differences.In the future,research on more general non-rigid differences algorithm will be considered,which has application value for using priori knowledge of MRI to reduce CT dose,or to improve the quality of CT reconstruction with normal dose.