Accuracy and reproducibility of co-registration techniques based on mutual information and normalized mutual information for MRI and SPECT brain images

A three-step, "brute-force" approach toward optimized affine spatial normalization

The first step in spatial normalization of magnetic resonance (MR) images commonly is an affine transformation, which may be vulnerable to image imperfections (such as inhomogeneities or "unusual" heads). Additionally, common software solutions use internal starting estimates to allow for a more efficient computation, which may pose a problem in datasets not conforming to these assumptions (such as those from children). In this technical note, three main questions were addressed: one, does the affine spatial normalization step implemented in SPM12 benefit from an initial inhomogeneity correction. Two, does using a complexity-reduced image version improve robustness when matching "unusual" images. And three, can a blind "brute-force" application of a wide range of parameter combinations improve the affine fit for unusual datasets in particular. A large database of 2081 image datasets was used, covering the full age range from birth to old age. All analyses were performed in Matlab. Results demonstrate that an initial removal of image inhomogeneities improved the affine fit particularly when more inhomogeneity was present. Further, using a complexity-reduced input image also improved the affine fit and was beneficial in younger children in particular. Finally, blindly exploring a very wide parameter space resulted in a better fit for the vast majority of subjects, but again particularly so in infants and young children. In summary, the suggested modifications were shown to improve the affine transformation in the large majority of datasets in general, and in children in particular. The changes can easily be implemented into SPM12.

The Supplementary Motor Area Responsible for Word Retrieval Decline After Acute Thalamic Stroke Revealed by Coupled SPECT and Near-Infrared Spectroscopy

Damage to the thalamus may affect cognition and language, but the underlying mechanism remains unknown. In particular, it remains a riddle why thalamic aphasia occasionally occurs and then mostly recovers to some degree. To explore the mechanism of the affected cognition and language, we used two neuroimaging techniques—single-photon emission computed tomography (SPECT), suitable for viewing the affected brain distribution after acute thalamic stroke, and functional near-infrared spectroscopy (f-NIRS), focusing on hemodynamic responses of the supplementary motor area (SMA) responsible for speech production in conjunction with the frontal aslant tract (FAT) pathway. SPECT yielded common perfusion abnormalities not only in the fronto–parieto–cerebellar loop, but also in the SMA, IFG and surrounding language-relevant regions. In NIRS sessions during a phonemic verbal fluency task, we found significant word retrieval decline in acute thalamic patients relative to age-matched healthy volunteers. Further, NIRS showed strong correlation between word retrieval and posterior SMA responses. In addition, follow-up NIRS exhibited increased bilateral SMA responses linked to improving word retrieval ability. The findings suggest that cognitive dysfunction may be related to the fronto–parieto–cerebellar loop, while language dysfunction is attributed to the SMA, IFG and language-related brain areas. SMA may contribute to the recovery of word retrieval difficulty and aphasia after thalamic stroke.

Eddy-current-induced distortion correction using maximum reconciled mutual information in diffusion MR imaging

PURPOSE

In diffusion tensor imaging, a large number of diffusion-weighted (DW) images with different diffusion gradient directions are attained during scanning. However, subjects' involuntary head movements and eddy current effect related to large diffusion-sensitizing gradients will cause distortions of DW images. Therefore, for tracking accurately white matter structures and tractography, the distortions have to be realigned before model fitting. Currently, traditional methods use maximum mutual information (MMI) or normalized mutual information (NMI) as similarity measure for DW images registration. These information measures are defined by Shannon entropy. The image entropy is able to embody the global information complexity but ignore the local information complexity caused by heterogeneous intensity contrasts in DW images, making registration algorithm early converge.

METHOD

To overcome the above problem, we present maximum reconciled mutual information (MRMI) combining both global information and local information as the similarity measure of the registration algorithm framework.

RESULT

(i) In comparison with traditional methods, under our proposed MRMI method, the border of DW image is more anastomotic with the b0 image, and the fitted fractional anisotropy (FA) map after registration is closer to the true brain boundary. (ii) By quantitative analysis of registration results, our method has a significant advantage over others in terms of NMI between b0 image and the aligned DW images.

CONCLUSION

The results suggest that there is a high-level matching in space between the b0 image and the DW images aligned by the MRMI method, raising the registration robustness and accuracy compared to the traditional DW registration methods. It may provide a better option for the existing diffusion image registration tools (e.g., FMRIB Software Library) and commonly multimodal medical image registration.

[Evaluation of the Image Registration Program for Portal Images Using CR and DRR Images in Radiation Therapy]

In this study, computer simulations and experiments were used to verify the accuracy of a two-dimensional image registration program (program) for portal images that we previously developed. The program used a computed radiography cassette system and digitally reconstructed radiography images as planning images for external beam radiation therapy. Using this program, we also investigated the reason two-dimensional automatic image registration images experienced large misregistration in clinical practice using commercial image registration systems. Mutual information and normalized mutual information were used as the registration criteria. To investigate the influence of image background with or without a region of interest (ROI), results of image registrations were compared. Parameters of image registration were defined as translation in the horizontal and vertical directions (x and y, respectively) and rotation (θ) around the axis perpendicular to the x-y plane. There was no significant difference in image registration arising from the difference between mutual information and normalized mutual information. Image registration was improved with a ROI. Regardless of the registration criteria, errors in image registration with a ROI in the experimental study were ≤1.2 mm in directions x and y and ≤1.0 degree in rotation θ. We found that image registration required setting up as close to the planned position as possible.

Operative simulation of anterior clinoidectomy using a rapid prototyping model molded by a three-dimensional printer

BACKGROUND

As the anatomical three-dimensional (3D) positional relationship around the anterior clinoid process (ACP) is complex, experience of many surgeries is necessary to understand anterior clinoidectomy (AC). We prepared a 3D synthetic image from computed tomographic angiography (CTA) and magnetic resonance imaging (MRI) data and a rapid prototyping (RP) model from the imaging data using a 3D printer. The objective of this study was to evaluate anatomical reproduction of the 3D synthetic image and intraosseous region after AC in the RP model. In addition, the usefulness of the RP model for operative simulation was investigated.

METHODS

The subjects were 51 patients who were examined by CTA and MRI before surgery. The size of the ACP, thickness and length of the optic nerve and artery, and intraosseous length after AC were measured in the 3D synthetic image and RP model, and reproducibility in the RP model was evaluated. In addition, 10 neurosurgeons performed AC in the completed RP models to investigate their usefulness for operative simulation.

RESULTS

The RP model reproduced the region in the vicinity of the ACP in the 3D synthetic image, including the intraosseous region, at a high accuracy. In addition, drilling of the RP model was a useful operative simulation method of AC.

CONCLUSIONS

The RP model of the vicinity of ACP, prepared using a 3D printer, showed favorable anatomical reproducibility, including reproduction of the intraosseous region. In addition, it was concluded that this RP model is useful as a surgical education tool for drilling.

Image coregistration: quantitative processing framework for the assessment of brain lesions

The quantitative, multiparametric assessment of brain lesions requires coregistering different parameters derived from MRI sequences. This will be followed by analysis of the voxel values of the ROI within the sequences and calculated parametric maps, and deriving multiparametric models to classify imaging data. There is a need for an intuitive, automated quantitative processing framework that is generalized and adaptable to different clinical and research questions. As such flexible frameworks have not been previously described, we proceeded to construct a quantitative post-processing framework with commonly available software components. Matlab was chosen as the programming/integration environment, and SPM was chosen as the coregistration component. Matlab routines were created to extract and concatenate the coregistration transforms, take the coregistered MRI sequences as inputs to the process, allow specification of the ROI, and store the voxel values to the database for statistical analysis. The functionality of the framework was validated using brain tumor MRI cases. The implementation of this quantitative post-processing framework enables intuitive creation of multiple parameters for each voxel, facilitating near real-time in-depth voxel-wise analysis. Our initial empirical evaluation of the framework is an increased usage of analysis requiring post-processing and increased number of simultaneous research activities by clinicians and researchers with non-technical backgrounds. We show that common software components can be utilized to implement an intuitive real-time quantitative post-processing framework, resulting in improved scalability and increased adoption of post-processing needed to answer important diagnostic questions.

Evaluation of delineation of image details in semiconductor PET utilizing the normalized mutual information technique

OBJECTIVE

PET using semiconductor detectors provides high-quality images of the human brain because of its high spatial resolution. To quantitatively evaluate the delineation of image details in clinical PET images, we used normalized mutual information (NMI) to quantify the similarity with images obtained through MRI. NMI is used to evaluate image quality by determining similarity with a reference image. The aim of this study was to evaluate quantitatively the delineation of image details provided by semiconductor PET.

MATERIALS AND METHODS

To quantitatively evaluate anatomical delineation in clinical PET images, MRI scans of patients were used as T1-weighted images. [(18)F]-fluorodeoxyglucose ((18)F-FDG) PET brain images were obtained from six patients using (a) a Hitachi semiconductor PET scanner and (b) a ECAT HR+ scintillator PET scanner. The NMI calculated from the semiconductor PET and MRI was denoted by NMIsemic, whereas the NMI calculated from conventional scintillator PET and MRI was denoted by NMIconve. The higher the value of NMI, the greater the similarity to MRI.

RESULTS

NMIsemic ranged from 1.22 to 1.29, whereas NMIconve ranged from 1.13 to 1.18 (P<0.05). Furthermore, all the NMI values of the semiconductor PET were higher than those of the conventional scintillator PET.

CONCLUSION

Utilizing NMI, we quantitatively evaluated the delineation of image details in clinical PET images. The results reveal that semiconductor PET has superior anatomical delineation and physical performance compared with conventional scintillator PET. This improved delineation of image details makes semiconductor PET promising for clinical applications.

Registration of Brain MRI/PET Images Based on Adaptive Combination of Intensity and Gradient Field Mutual Information

Traditional mutual information (MI) function aligns two multimodality images with intensity information, lacking spatial information, so that it usually presents many local maxima that can lead to inaccurate registration. Our paper proposes an algorithm of adaptive combination of intensity and gradient field mutual information (ACMI). Gradient code maps (GCM) are constructed by coding gradient field information of corresponding original images. The gradient field MI, calculated from GCMs, can provide complementary properties to intensity MI. ACMI combines intensity MI and gradient field MI with a nonlinear weight function, which can automatically adjust the proportion between two types MI in combination to improve registration. Experimental results demonstrate that ACMI outperforms the traditional MI and it is much less sensitive to reduced resolution or overlap of images.

Quantitative multi-compartmental SPECT image analysis for lateralization of temporal lobe epilepsy

This study assesses the utility of compartmental analysis of SPECT data in lateralizing ictal onset in cases of a putative mesial temporal lobe epilepsy (mTLE). An institutional archival review provided 46 patients (18M, 28F) operated for a putative mTLE who achieved an Engel class Ia postoperative outcome. This established the standard to assure a true ictal origin. Ictal and interictal SPECT images were separately coregistered to T1-weighted (T1W) magnetic resonance (MR) image using a rigid transformation and the intensities matched with an l(1) norm minimization technique. The T1W MR image was segmented into separate structures using an atlas-based automatic segmentation technique with the hippocampi manually segmented to improve accuracy. Mean ictal-interictal intensity difference values were calculated for select subcortical structures and the accuracy of lateralization evaluated using a linear classifier. Hippocampal SPECT analysis yielded the highest lateralization accuracy (91%) followed by the amygdala (87%), putamen (67%) and thalamus (61%). Comparative FLAIR and volumetric analyses yielded 89% and 78% accuracies, respectively. A multi-modality analysis did not generate a higher accuracy (89%). A quantitative anatomically compartmented approach to SPECT analysis yields a particularly high lateralization accuracy in the case of mTLE comparable to that of quantitative FLAIR MR imaging. Hippocampal segmentation in this regard correlates well with ictal origin and shows good reliability in the preoperative analysis.

Comparative study of anatomical normalization errors in SPM and 3D-SSP using digital brain phantom

OBJECTIVE

In single photon emission computed tomography (SPECT) cerebral blood flow studies, two major algorithms are widely used statistical parametric mapping (SPM) and three-dimensional stereotactic surface projections (3D-SSP). The aim of this study is to compare an SPM algorithm-based easy Z score imaging system (eZIS) and a 3D-SSP system in the errors of anatomical standardization using 3D-digital brain phantom images.

METHODS

We developed a 3D-brain digital phantom based on MR images to simulate the effects of head tilt, perfusion defective region size, and count value reduction rate on the SPECT images. This digital phantom was used to compare the errors of anatomical standardization by the eZIS and the 3D-SSP algorithms.

RESULTS

While the eZIS allowed accurate standardization of the images of the phantom simulating a head in rotation, lateroflexion, anteflexion, or retroflexion without angle dependency, the standardization by 3D-SSP was not accurate enough at approximately 25° or more head tilt. When the simulated head contained perfusion defective regions, one of the 3D-SSP images showed an error of 6.9% from the true value. Meanwhile, one of the eZIS images showed an error as large as 63.4%, revealing a significant underestimation.

CONCLUSION

When required to evaluate regions with decreased perfusion due to such causes as hemodynamic cerebral ischemia, the 3D-SSP is desirable. In a statistical image analysis, we must reconfirm the image after anatomical standardization by all means.

An automatic MRI/SPECT registration algorithm using image intensity and anatomical feature as matching characters: application on the evaluation of Parkinson's disease

Single-photon emission computed tomography (SPECT) of dopamine transporters with (99m)Tc-TRODAT-1 has recently been proposed to offer valuable information in assessing the functionality of dopaminergic systems. Magnetic resonance imaging (MRI) and SPECT imaging are important in the noninvasive examination of dopamine concentration in vivo. Therefore, this investigation presents an automated MRI/SPECT image registration algorithm based on a new similarity metric. This similarity metric combines anatomical features that are characterized by specific binding, the mean count per voxel in putamens and caudate nuclei, and the distribution of image intensity that is characterized by normalized mutual information (NMI). A preprocess, a novel two-cluster SPECT normalization algorithm, is also presented for MRI/SPECT registration. Clinical MRI/SPECT data from 18 healthy subjects and 13 Parkinson's disease (PD) patients are involved to validate the performance of the proposed algorithms. An appropriate color map, such as "rainbow," for image display enables the two-cluster SPECT normalization algorithm to provide clinically meaningful visual contrast. The proposed registration scheme reduces target registration error from >7 mm for conventional registration algorithm based on NMI to approximately 4 mm. The error in the specific/nonspecific (99m)Tc-TRODAT-1 binding ratio, which is employed as a quantitative measure of TRODAT receptor binding, is also reduced from 0.45+/-0.22 to 0.08+/-0.06 among healthy subjects and from 0.28+/-0.18 to 0.12+/-0.09 among PD patients.

Improved accuracy of brain MRI/SPECT registration using a two-cluster SPECT normalization algorithm and a combinative similarity measure: application to the evaluation of Parkinson's disease

OBJECTIVE

Single-photon emission computed tomography (SPECT) of dopamine transporters with technetium-99-labeled tropane derivative (99m)Tc-TRODAT-1 has recently been suggested to offer valuable information in assessing the functionality of dopaminergic systems. To facilitate the non-invasive examination of the dopamine concentration in vivo, registering magnetic resonance imaging (MRI) and SPECT image is important. This article proposes a new similarity measure for MRI/SPECT registration.

METHODS

The proposed similarity measure combines anatomic features that are characterized by specific binding of nuclear medicine and the distribution of image intensity that are characterized by the normalized mutual information (NMI). A preprocess, a novel two-cluster SPECT normalization algorithm, is also proposed.

RESULTS

Compared with the conventional NMI-based registration algorithm, the proposed registration framework reduces the target of registration error from >7 mm to approximately 4 mm. The error of the specific-to-non-specific (99m)Tc-TRODAT-1 binding ratio (BR), a quantitative measure of TRODAT receptor binding, is also reduced from 0.45 to 0.08 in the healthy subjects and from 0.28 to 0.12 in Parkinson's disease patients.

CONCLUSIONS

A suitable color map, such as "rainbow," for image display enables the two-cluster SPECT normalization algorithm to provide clinically meaningful visual contrast. In addition, registering MRI/SPECT based on the proposed similarity measure improves the accuracy compared with the conventional NMI-based algorithm.

Comparison of fMRI coregistration results between human experts and software solutions in patients and healthy subjects

Functional magnetic resonance imaging (fMRI) performed by echo-planar imaging (EPI) is often highly distorted, and it is therefore necessary to coregister the functional to undistorted anatomical images, especially for clinical applications. This pilot study provides an evaluation of human and automatic coregistration results in the human motor cortex of normal and pathological brains. Ten healthy right-handed subjects and ten right-handed patients performed simple right hand movements during fMRI. A reference point chosen at a characteristic anatomical location within the fMRI sensorimotor activations was transferred to the high resolution anatomical MRI images by three human fMRI experts and by three automatic coregistration programs. The 3D distance between the median localizations of experts and programs was calculated and compared between patients and healthy subjects. Results show that fMRI localization on anatomical images was better with the experts than software in 70% of the cases and that software performance was worse for patients than healthy subjects (unpaired t-test: P = 0.040). With 45.6 mm the maximum disagreement between experts and software was quite large. The inter-rater consistency was better for the fMRI experts compared to the coregistration programs (ANOVA: P = 0.003). We conclude that results of automatic coregistration should be evaluated carefully, especially in case of clinical application.

Simultaneous surface registration of ictal and interictal SPECT and magnetic resonance images for epilepsy studies

BACKGROUND

Subtraction of ictal and interictal single photon emission computed tomography (SPECT) images is known to be successful in localizing the seizure focus in the pre-surgical evaluation of patients with partial epilepsy. A computer-aided methods for producing subtraction ictal SPECT co-registered to the magnetic resonance image (MRI) (the SISCOM method) is commonly used. The two registrations involved in SISCOM are (1) between the ictal-interictal SPECT images, which was shown to be the more critical, and (2) between the ictal image and MRI.

OBJECTIVE

To improve the accuracy of ictal-interictal registration in SISCOM by registering all three images (ictal, interictal SPECT, MRI) simultaneously.

METHODS

The registration problem is formulated as the minimization of a cost function between three surfaces. Then, to achieve a global minimum of this cost function, the Powell algorithm with randomly distributed initial configurations is used. This technique is tested by a realistic simulation study, a phantom study and a patient study.

RESULTS

The results of the simulation study demonstrate that, in surface-based registration, the triple-registration method results in a smaller ictal-interictal SPECT registration error than the pair-wise registration method (P<0.05) for a range of values of the cost-function parameter. However, the improved registration error is still larger than that obtained by the normalized mutual information method (P<0.001), which is a voxel-based registration algorithm. The phantom and patient studies reveal no observable difference between registration results.

CONCLUSIONS

Although the improved accuracy of triple registration is slightly worse than voxel-based registration, it will soon be possible to apply the results of this study in research utilizing the triple-registration principle to improving voxel-based results of ictal-interictal registration.