ABSTRACT
Non-rigid registration plays an important role in medical image analysis. U-Net has been proven to be a hot research topic in medical image analysis and is widely used in medical image registration. However, existing registration models based on U-Net and its variants lack sufficient learning ability when dealing with complex deformations, and do not fully utilize multi-scale contextual information, resulting insufficient registration accuracy. To address this issue, a non-rigid registration algorithm for X-ray images based on deformable convolution and multi-scale feature focusing module was proposed. First, it used residual deformable convolution to replace the standard convolution of the original U-Net to enhance the expression ability of registration network for image geometric deformations. Then, stride convolution was used to replace the pooling operation of the downsampling operation to alleviate feature loss caused by continuous pooling. In addition, a multi-scale feature focusing module was introduced to the bridging layer in the encoding and decoding structure to improve the network model's ability of integrating global contextual information. Theoretical analysis and experimental results both showed that the proposed registration algorithm could focus on multi-scale contextual information, handle medical images with complex deformations, and improve the registration accuracy. It is suitable for non-rigid registration of chest X-ray images.
Subject(s)
Algorithms , Learning , ThoraxABSTRACT
Objective@#To evaluate the precision of image registration between MRI simulation (MRIsim) and CT simulation compared to diagnostic MRI(MRIdiag) and to provide information for further application of MRIsim.@*Methods@#A total of 24 patients who underwent both MRIsim and MRIdiag were enrolled, including 8 patients with gliomas, 8 with nasopharyngeal carcinomas and 8 with prostate cancers. MRIsim and MRIdiag images of each patient were fused with CT. The OARs were delineated on three modalities of images and targets were delineated on fusion image of MRIsim with CT (F_CTMsim) and fusion image of MRIdiag with CT (F_CTMdiag) respectively. The concordance index (CI), Dice′s similarity coefficient (DSC) between the OARs and image similarity index (S) based on images from MRIsim, MRIdiag and CT were evaluated. IMRT plans were designed based targets on F_CTMsim and OARs on CT images, and differences in dosimetry of targets and OARs were evaluated subsequently.@*Results@#Volumes of most OAR from three modalities of images showed no statistically significant difference(P>0.05). All the CI and DSC between the OARs derived from MRIsim and CT were higher than those corresponding values from MRIdiag, and a statistically significant difference was achieved in 50% of these OARs (t=2.58-5.47, P<0.05). The S values of MRIsim and MRIdiag compared with CT were 0.89 and 0.83 respectively (t=5.77, P<0.05). MRIsim improved the S value by 10% (2%-56%) compared with MRIdiag. No further differences in dosimetry were found on all OARs and all targets(P>0.05).@*Conclusions@#The precision of image registration can be significantly improved by introducing MRIsim into radiotherapy planning design compared with MRIdiag. However, no significant differences in dosimetry were found on targets produced by rigid registration and manual adjustment method .
ABSTRACT
Objective: To summarize the major progress in medical image registration in recent years. Furthermore, based on the recent advances in this field, this paper can provide a reference in following domains: three-dimensional medical image reconstruction, medical image visualization, quantitative analysis. Methods: Firstly, referring to a myriad of latest papers on medical image registration. Secondly, analyzing traits and exiting problems of techniques which presented in those papers. Finally, putting forward some efficient methods for solving these problems. Results: This paper compares the characteristics of some typical algorithms and its application and looks forward to the future research work. Conclusion: Using optimization strategy to improve the quality of image registration and studying on non-rigid image registration are the directions for future research in medical image registration field.
ABSTRACT
Objective To register two breath-hold lung volumes image from one subject with deep expiration and deep inspiration.Methods Three pairs of thoracic high resolution CT serial from three subjects were collected under two breath-hold respiration stages.The lung parenchyma of every serial was segmented using the serial segmentation algorithm.Left and right lungs were stored separately.Expiration and inspiration volume images of single lung were registered.Firstly,affine transformation parameters were found based on the anatomic flag surfaces and expiration image volume was re-sampled with affine transformation.Secondly,"Demons" algorithm was employed to register two image volumes non-rigidly.Results Two lung surfaces and the inner structures have a nice registration.The average volume overlap of two images before registration is 0.7982.After global affine transformation,it improves to 0.8936.After "Demons",it is up to 0.9544.The average descending percentage of root mean square errors is 19.83%(after the global affine transformation) and 49.43%(after the "Demons" non-rigid registration).Conclusion The intra-subject registration between two lung image volumes with large deformations described here has an effective registration result.It offers a good base to analyze the lung respiration function.
ABSTRACT
Objective To register pre-operative MRI/CT images with intra-operative ultrasound images based on vessels visible in both of the modalities. Method A non-rigid registration method of multimodal medical images based on Free Form Deformation(FFD) was proposed. When the images were aligned, the centerline points of the vessels in one image aligned with the intensity ridge points in the other image. Rigid transformation was adopted in global registration while local deformation was described by a Free Form Deformation based on a modally controlled B-spline. The method applied an optimization strategy combining the genetic algorithm with the conjugated gradients algorithm to minimize the objective function. Result Two experiments were designed on phantom and clinical data to evaluate the method. The results demonstrated that the registration method was consistent accurate. The average standard deviation of the final transformation parameters was sub-voxel, sub-millimeter, and within 0.010 radians. Conclusion The results show that the method has good registration accuracy and convergence rate. It can be applied efficiently in the ultrasound-image-guided surgery system.
ABSTRACT
Objective To present a new algorithm for multidimensional medical image registration from global registration to local registration in sequence. Methods Firstly, the global registration was achieved by the method of affine transformation composed of B-splines,whose knots were the four vertexes of the medical image. Then the knots of the B-splines were increased, and the transformation function was more complex and elastic than ever,which completed the elastic aligning for the detail of the medical image. Results The whole registration algorithm represented the principle aligning from global registration to local registration. Conclusion It is proved by experiments that the presented algorithm can decrease the time of calculation and increase the robustness of registration.