Medical image enhancement based on volumetric tissue segmentation fusion (Uni-stable 3D method)

doi:10.21203/rs.3.rs-2132692/v1

Download PDF

Research Article

Medical image enhancement based on volumetric tissue segmentation fusion (Uni-stable 3D method)

https://doi.org/10.21203/rs.3.rs-2132692/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

The 3D Uni-stable method is a novel technique for 3D medical image enhancement which produces 3D Images of high contrast from the scanned anisotropic scaling images. This is done by estimating some intermediate slices through resizing the original scans. Rescaling has been achieved at three different levels: rescaling of eigenvalues of diffusion, rescaling the Scalar Indexes from the original eigenvalues, and rescaling the cluster maps of the segmentation of the original Scalar Indexes. Four interpolation methods have been employed at each level and four clustering algorithms have been employed in the process. The 3D Uni-stable image is almost universal as it combines variety of algorithms points of views into one 3D probability map. This reduces boundary-overlapping among different tissues, and hence improves the uniqueness of the segmentation problem solution. The stability factor of the 3D Uni-stable-Images is measured by maximum match analysis between the cluster maps which are generated from 3D Uni-stable images using variety of clustering methods with respect to true fact references for 5 different brains and the resultant standard deviations of Uni-stable images maximum match analysis in both threshold and tissue to brain ratio are much lower than Mean Diffusivity and Fractional Anisotropy scalar indexes for both CSF/non-CSF and WM/non-WM respectively.

Uni-stable

CSF

Segmentation

Brain

Computer-aided diagnosis (CAD) is considered one of the major research subjects in medical imaging and diagnostic radiology [1, 2]. With CAD, radiologists use the computer output as a "second opinion" and make the final decisions. The ability to produce high contrast medical images has its importance in diagnosing, evaluation and computer based simulation for a certain tissue or organ.

Medical imaging such as CT-scan and MRI are usually accomplished in anisotropic manner where the planer resolution spacing and slice to slice spacing are (usually) different [3]. This makes the 2D based segmentation process of scans not able to provide the accurate 3D bounds of the tissues, where this show the necessity for estimating some intermediate plans to solve that problem. The estimation of some intermediate plans requires 3D image resizing to be involved in the process.

In addition to the anisotropic spacing problem, it is only possible to obtain macroscale images because of scanners’ limited resolution, where each sample represents the average property of the corresponding group of tissues for each zone [4, 5]. This limitation is the main reason for boundary-overlapping problem between neighbor tissues in medical images [6–8], and as a result, automatic medical image segmentation is difficult to be achieved. With those 2 problems, the estimation of accurate 3D tissue’s boundary becomes in demand.

When only 2D images of the original scans are employed for processing of tissue in 3D, it would act like resizing this image using nearest interpolation method. However, by applying other interpolation methods such as Linear, Cubic, Lanczos2 or Lanczos3 [9, 10], more accurate estimation for the intermediate plans would be expected. Linear interpolation estimates intermediate states over a tangent line between two measured states while Cubic uses Gaussian smoothing curve and Lanczos methods depends on the Sinc function.

The most important image segmentation techniques are: threshold based methods, clustering based methods, edge based methods, region based methods, watershed based methods, Partial Differential Equation (PDE) based methods and Artificial Neural Network (ANN) based methods [11]. Clustering base methods such as Otsu [12], K-Means [13], Expectation Maximization (EM) [14], C-Means, Fuzzy C-Means (FCM) and Fuzzy C-Means algorithm with spatial constraints (SFCM) [15], Markov Random Field (MRF) and Iterated Conditional Modes (ICM) [16] are widely used with medical images.

It was reported that there is no unique solution for the segmentation problem, because of the effect of the used imaging modality on the segmentation process; different results are obtained by changing clustering method and/or the selected numbers of clusters [5, 7, 8, 15, 16]. One solution in a former research [17] has been developed to produce Uni-stable images where segmentation results are relatively stable regarding the change of the clustering method. However, this was tested for the 2D images and need to be improved to cover the 3D images as well.

Human brain is taken as case study for this research. Brain is composed of three main components: White Matter (WM), Gray Matter (GM) and Cerebrospinal Fluid (CSF). CSF contains the largest amount of water among WM and GM. Arrangement of water particles for each of these components differs as water particles of CSF are completely free; however, water contents of WM are arranged in very restricted order inside the myelinated axons and for GM, water particles are arranged in axons but not very restricted as in WM. In other terms, CSF is isotropic material with larger amount of water particles, GM is anisotropic material with relatively low amount of water and finally WM is highly-anisotropic material with relatively the lowest amount of water. Among different medical imaging modalities, DTI is capable of providing more information about different tissues structures, and hence the ability to segmenting these structures from each other for further processing such as diagnosing and simulation.

Many models have been represented for 3D brain segmentation. Duy M. H. Nguyen et. al. produces a 3D brain tissues segmentation using Gaussian Mixture Models (GMMs), Convolution Neural Networks (CNNs) and Deep Neural Networks (DNNs)[18]. Yuankai Huo et.al. proposed the Spatially Localized Atlas Network Tiles (SLANT) method to distribute multiple independent 3D Fully Convolutional Networks (FCN) for high-resolution whole brain segmentation[19]. Fareeen Ramzan et.al. employ a network for the segmentation of multiple brain regions has been proposed that is based on 3D convolutional neural networks and utilizes residual learning and dilated convolution operations to efciently learn the end-to-end mapping from MRI volumes to the voxel-level brain segments[20]. Youyong Kong et.al. presents a novel algorithm for brain tissue segmentation based on supervoxel and graph filter. Firstly, an effective supervoxel method is employed to generate effective supervoxels for the 3D MRI image. Secondly, the supervoxels are classified into different types of tissues based on filtering of graph signals [21]. Kamarujjaman represnt a novel 3D unsupervised spatial fuzzy-based brain MRI volume segmentation technique in the presence of intensity inhomogeneity and noise. Instead of static masking, dynamic 3D masking has been proposed to measure the correlation among neighbors [22]. Results from stated methods targeting either high resolution scans or depends on one segmentation technique. The main aspect of this research is to generate 3D images which are universal and stable for segmentation process.

2.1. Diffusion Tensor Imaging

Diffusion Tensor Imaging (DTI) is an MRI technique that enables the measurement of the restricted diffusion of water particles in tissue [4]. DTI produces data that describes tissues fibers in terms of eigenvalues and eigenvectors. For medical applications and diagnosing therapies, DTI data are usually described as Grayscale images in terms of rotational invariant quantities or Scalar Indices (SI) such as Mean Diffusivity (MD), Fractional Anisotropy (FA), Rational Anisotropy (RA), Linear Anisotropy (Cl), Spherical Anisotropy (Cs), Volume Ratio (VR), Angular Anisotropy (AA), and Diffusion Volume (DV) [4, 5, 6, 23–27]. Datasets of this study have been acquired from John Hopkins Medical Institute web site. These datasets consists of brains’ DTI volume data. Each brain DTI volume contains 50 slices of 256 × 256 voxels per slice. Each dataset has a file that contains 35 gradient orientations which are used to calculate tensor data. In every slice, voxel width = 0.9375 mm, voxel height = 0.9375 mm and space between two successive slices = 2.5mm.

2.2 3D Uni-stable Images

The lake of unique solution for solving medical images segmentation problem is caused by the wide range of selections among different medical imaging modalities, clustering methods, interpolation method of the rescaling its order where each setting has its own estimate. A new approach for 3D brain image segmentation based on interpolating anisotropic brain scans from Diffusion Tensor Imaging modality. As axial brain scans are usually accomplished in anisotropic spacing where the plan resolution spacing and slice to slice spacing are usually different, then 2D based segmentation of brain tissues cannot provide accurate 3D bounds of the tissues. This is where 3D image interpolation should be involved in the process. The main concerns of the accurate 3D brain tissues are the suitable stage order of the interpolation in the process, the suitable interpolation methods and the suitable Scalar indexes for that job.

Uni-stable method [17] has been originally developed for generating high contrast 2D images based on DTI scans, where the Uni-stable method has been proved to be universal, where all estimates of different regular segmentation settings are considered in the solution, and they are relatively stable, because results of their segmentation are proved to be relatively independent of the applied clustering method (Fig. 1). The process starts by producing different SI from eigenvalues of the DTI data, generate Cluster Maps (CM) from the generated SIs and finally fuse the generated CMs into one probability map which is the Uni-stable Image where this process makes sure that common points will have more probability than less common points.

In the way to upgrade this method to 3D version, resizing stage would be added to estimate some intermediate slices between originally scanned ones which in turn solve the anisotropic spacing problem. The selected interpolation method of the resize operation and its order in the process plays a basic role that affect the 3D Uni-stable image generation as shown in Fig. 2. The first setting (Setting 1) is resizing the eigenvalues of the DTI images before the other stages. The 2nd (Setting 2) is to generate SI’s from the original eigenvalues and then resize the SI maps. The last one (Setting 3) is to generate the CM’s from original SIs and then resize them before the fusion process.

Four interpolation methods are tested for 3D image resizing: Linear, Cubic, Lanczos2 and Lanczos3. For N SIs, which contain clear features about the required tissues; M clustering methods with P different Interpolation methods generate N × M × P estimates of normalized CMs. Normalized Uni-stable-Image IMG_Uni is generated by fusing of these CMs together, where:

IMG_Uni. = \(\frac{\sum _{n=1}^{N}\sum _{m=1}^{M}\sum _{p=1}^{P}{CM}_{nmp}}{Max\left(\sum _{n=1}^{N}\sum _{m=1}^{M}\sum _{p=1}^{P}{CM}_{nmp}\right)}\) ……………………...…………… (1)

In this study Uni-stable method is applied to brain’s DTI’s Scalar Indices to distinguish between CSF and non-CSF tissues and again to distinguish between WM and non-WM tissues. In both cases, four clustering methods are used to generate CMs, namely: Otsu, KM, FCM and SFCM. The generic setting for the Uni-stable method is chosen by using only CMs which provides estimates of 3 clusters, namely: tissue, non-tissue and background. Brain data has been resized from 256x256x50 to 256x256x200 using the three different settings to estimate some intermediate slices between originally scanned ones.

2.3. CSF/non-CSF 3D Uni-stable-Image generation

Five SIs are used to generate CSF/non-CFS Uni-stable-Images (CSF_Uni), where CSF appears in these images as light regions; and they are: MD, DV, λ1, λ2, and λ3. Those SI has been clustered using 4 different clustering algorithms namely: Otsu, KM, FCM and SFCM. Rescaling using four interpolation methods has been tested in the process with the different location. For each setting, 80 different cluster maps are produced. Cluster maps are then fused into one image as shown in Fig. 3.

Table 1

Different Settings for CSF/non-CSF 3D Uni-stable Image Generation.
	Rescale Level	Interpolation Methods	Clustering Methods	Scalar Indexes	Number of Fused Cluster Maps
Setting 1	Eigenvalues Rescale	Linear, Cubic, Lanczos2, Lanczos3	Otsu, K-Means, FCM, SFCM	MD, DV, λ₁, λ₂, λ₃	80
Setting 2	Scalar Indexes Rescale	Linear, Cubic, Lanczos2, Lanczos3	Otsu, K-Means, FCM, SFCM	MD, DV, λ₁, λ₂, λ₃	80
Setting 3	Cluster Maps Rescale	Linear, Cubic, Lanczos2, Lanczos3	Otsu, K-Means, FCM, SFCM	MD, DV, λ₁, λ₂, λ₃	80

2.4. WM/non-WM 3D Uni-stable-Image generation

Five SIs are used to generate WM/non-WM Uni-stable-Images (WM_Uni), where WM appears in these images as light regions; and they are: FA, RA, AA, MA, and Cl. Those SI has been clustered using 4 different clustering algorithms namely: Otsu, KM, FCM and SFCM. Rescaling using four interpolation methods has been tested in the process with the different location. For setting 1, 40 different cluster maps are produced because only two interpolation methods re used and 80 different cluster maps are use in other settings. More justification would be indicated in the discussion and conclusion part about that. Finally, cluster maps are then fused into one image as shown in Fig. 4.

Table 2

Different Settings for WM/non-WM 3D Uni-stable Image Generation.
	Rescale Level	Interpolation Methods	Clustering Methods	Scalar Indexes	Number of Fused Cluster Maps
Setting 1	Eigenvalues Rescale	Linear, Cubic	Otsu, K-Means, FCM, SFCM	FA, RA, AA, MA, Cl	40
Setting 2	Scalar Indexes Rescale	Linear, Cubic, Lanczos2, Lanczos3	Otsu, K-Means, FCM, SFCM	FA, RA, AA, MA, Cl	80
Setting 3	Cluster Maps Rescale	Linear, Cubic, Lanczos2, Lanczos3	Otsu, K-Means, FCM, SFCM	FA, RA, AA, MA, Cl	80

3.1 Reference Model

The usual SIs which are used in the brain image segmentation are FA and MD. It was reported that thresholds of normalized MD images with values around 0.3 can efficiently separate the images into CSF/non-CSF clustering maps [28]. Furthermore, thresholds of normalized FA images with values between 0.25 and 0.45 can separate images into WM/non-WM clustering maps [29]. As original scans reflect the solid fact about the scans so they should be taken as comparison reference. For size consistency, nearest interpolation is be applied to resize the original FA and MD.

3.2 Non-Uni-stable 3D images Segmentation

For the regular process, FA and MD are resized using linear interpolation or cubic interpolation after being calculated from the original eigenvalues [30]. However more setting has been explored where FA and MD are resized according to the following tables:

Table 3

Different Settings for MD rescaling.
	Rescale Level	Interpolation Methods
MD1.1	Eigenvalues Rescale	Linear
MD1.2	Eigenvalues Rescale	Cubic
MD2.1	Scalar Indexes Rescale	Linear
MD2.2	Scalar Indexes Rescale	Cubic

Table 4

Different Settings for FA rescaling.
	Rescale Level	Interpolation Methods
FA1.1	Eigenvalues Rescale	Linear
FA1.2	Eigenvalues Rescale	Cubic
FA2.1	Scalar Indexes Rescale	Linear
FA2.2	Scalar Indexes Rescale	Cubic

And the four mentioned clustering methods are also applied on those images with maximum match as shown in Fig. 5 and Fig. 6. It is noticed that each clustering method has different threshold point and this is the normal case where non unique solution is obtained.

Referring to Maximum-Match analysis for 5 brains, the average mean and the average standard deviation of the clustering algorithm and regular threshold method and the CSF to Brain and WM to Brain ratios are given in Tables 5 and Table 6

Table 5

Average Mean and Average standard deviation of 5 human brains for maximum match Threshold value and CSF to Brain ratio based on MD Images of different settings
	Threshold	CSF to Brain Ratio
MD1.1	0.306 ± 0.0118	18.75 ± 1.07
MD1.2	0.306 ± 0.0125	18.64 ± 1.06
MD2.1	0.306 ± 0.0118	18.75 ± 1.07
MD2.2	0.306 ± 0.0125	18.64 ± 1.06

Table 6

Average Mean and Average standard deviation of 5 human brains for maximum match Threshold value and WM to Brain ratio based on FA Images of different settings
	Threshold	WM to Brain Ratio
FA1.1	0.371 ± 0.0127	29.15 ± 1.80
FA1.2	0.375 ± 0.0142	28.57 ± 1.92
FA2.1	0.355 ± 0.0203	32.71 ± 8.58
FA2.2	0.368 ± 0.0109	28.26 ± 1.56

3.3 Uni-stable 3D images Segmentation

Maximum-Match analysis for 5 brains is repeated again for 3D Universal images under the 3 settings (Fig. 7, Fig. 8, Table 7 and Table 8). In both Fig. 7 and Fig. 8: Otsu, K-Means and FCM curves are overlapping. The SFCM is shifted a little bit downward below them; however, all of them are aligned almost to the same threshold point. This indicates that segmentation results are almost independent from the clustering algorithm.

Quantitatively, the average mean and average standard deviation of 5 human brains for maximum match Threshold value and CSF to Brain ratio based on 3D Universal images of both CSF/non-CSF and WM/non-WM under the different three settings are demonstrated in Table 7 and Table 8 respectively. The average mean values are almost the same such as Table 5 and Table 6; however average standard deviation values are much lower than the MD and FA cases. This indicates that segmentation of the 3D Uni-stable images almost independent of the clustering method.

Table 7

Average Mean and Average standard deviation of 5 human brains for maximum match Threshold value and CSF to Brain ratio based on 3D Universal CSF/non-CSF Images of different settings
	Threshold	CSF to Brain Ratio
CSF_Uni1	0.310 + 0.0050	18.44 + 0.47
CSF_Uni2	0.310 + 0.0050	18.46 + 0.40
CSF_Uni3	0.305 + 0.0051	17.99 + 0.34

Table 8

Average Mean and Average standard deviation of 5 human brains for maximum match Threshold value and WM to Brain ratio based on 3D Universal WM/non-WM Images of different settings
	Threshold	WM to Brain Ratio
WM_Uni1	0.387 + 0.0030	27.09 + 0.49
WM_Uni2	0.383 + 0.0019	26.41 + 0.38
WM_Uni3	0.380 + 0.0102	27.45 + 1.08

For CSF/non-CSF 3D Uni-stable images Setting 3 provides the lowest average standard deviations and hence the most stable setup. For WM/non-WM 3D Uni-stable images Setting 2 provides the most stable setup.

The 3D Uni-stable method is a novel technique for 3D medical image enhancement which produces 3D Images of high contrast from the scanned anisotropic scaling images. It is basically an upgraded version to the former Uni-stable method that handles 2D images. This is done by estimating some intermediate slices through resizing the original scans. Rescaling has been achieved at three different levels: rescaling of eigenvalues of diffusion, rescaling the Scalar Indexes from the original eigenvalues, and rescaling the cluster maps of the segmentation of the original Scalar Indexes. Four interpolation methods have been employed at each level and four clustering algorithms have been employed in the process.

This make 3D Uni-stable image is almost universal as it combines variety of algorithms points of views into one 3D probability map. This reduces boundary-overlapping among different tissues, and hence improves the uniqueness of the segmentation problem solution.

The stability factor of the 3D Uni-stable-Images is measured by maximum match analysis between the cluster maps which are generated from 3D Uni-stable images using variety of clustering methods with respect to true fact references. For CSF/non-CSF true fact reference is taken as threshold of MD scalar index’s histogram after being resized using nearest interpolation and again, for WM/non-WM true fact reference is taken as threshold of FA scalar index’s histogram after being resized using nearest interpolation. The same is repeated again but using cluster maps which are generated by applying some clustering methods to MD and FA. The resultant standard deviations of Uni-stable images maximum match analysis in both threshold and tissue to brain ratio are much lower than MD and FA. This reflects the stability of the Uni-stable images and indicates that they are almost clustering-algorithm-independent. The same process is repeated for 5 brains and results were also the same

Cubic, Lanczos2 and Lanczos 3 produce some negative values because of their kernel function, which can be eliminated. However, Lanczos2 and Lanczos3 Interpolation methods might not be suitable to be used in the level of resizing the eigenvalues because they depend on Sinc function. As Sinc function oscillates, some small positive values could be produced in the background zone. Because of the rational nature of WM/non-WM scalar indexes calculations, extra regions would be generated in the background zone such as FA images (Fig. 9). However, for integration based scalar indexes which are used to find CSF/non-CSF scalar indexes such as MD, these small values have almost no effect (Fig. 9).

Finally, it is concluded that the 3D Uni-stable method might be a straightforward solution for the segmentation problem of anisotropic scaling medical images, where the segmentation process can be achieved using one 3D Uni-stable image and any clustering method. For CSF/non-CSF, setting 3 produces the best results and for WM/non-WM, setting 2 produces best results.

CAD	Computer-aided diagnosis
PDE	Partial Differential Equation
ANN	Artificial Neural Network
EM	Expectation Maximization
FCM	Fuzzy C-Means
SFCM	Spatial Fuzzy C-Means
MRF	Markov Random Field
ICM	Iterated Conditional Modes
WM	White Matter
GM	Gray Matter
CSF	Cerebrospinal Fluid
GMMs	Gaussian Mixture Models
CNNs	Convolution Neural Networks
DNNs	Deep Neural Networks
SLANT	Spatially Localized Atlas Network Tiles
FCN	Fully Convolutional Networks
DTI	Diffusion Tensor Imaging
SI	Scalar Indix
MD	Mean Diffusivity
FA	Fractional Anisotropy
RA	Rational Anisotropy
Cl	Linear Anisotropy
Cs	Spherical Anisotropy
VR	Volume Ratio
AA	Angular Anisotropy
DV	Diffusion Volume
CM	Cluster Maps
MRI	Magnetic Resonance Imaging
CSFUni	CSF/non-CSF Uni-stable Image
WMUni	WM/non-WM Uni-stable Image

Acknowledgements

Laboratory of Brain Anatomical MRI – Johns Hopkins Medical Institute at Johns Hopkins University for provision of DT-MRI data and DTI-Studio utility.

Data Availability

Data is available in the website of the Laboratory of Brain Anatomical MRI – Johns Hopkins Medical Institute at Johns Hopkins University https://cmrm.med.jhmi.edu/

Doi K. Computer-aided diagnosis in medical imaging: Historical review, current status and future potential. Computerized Medical Imaging and Graphics. 2017; 31(4):198-211. DOI: 10.1016/j.compmedimag.2007.02.002
Li Q, Nishikawa RM, editors. Computer-Aided Detection and Diagnosis in Medical Imaging. Taylor & Francis, CRC Press, New York; 2015.
Neubert A, Salvado O, Acosta O, Bourgeat P, Fripp J. Constrained reverse diffusion for thick slice interpolation of 3D volumetric MRI images. Computerized Medical Imaging and Graphics 2012; 36(2): 130-138. DOI: 10.1016/j.compmedimag.2011.08.004.
Mori S. Introduction to Diffusion Tensor Imaging.1st ed., Oxford, UK: Elsevier, 2007.
Liu T, Li H, Wong K, Tarokh A, Guo L, Wong S. Brain Tissue Segmentation Based on DTI Data. Neuroimage 2007; 38: 114-23. DOI: 10.1016/j.neuroimage.2007.07.002
Zarei M, Johansen Berg H, Matthews PM. Diffusion Tensor Imaging and Tractography in Clinical Neurosciences. Iran J Radiol 2003; 1: 45-52.
Khotanlou H, Colliot O, Atif J, Bloch I. 3D Brain Tumor Segmentation in MRI Using Fuzzy Classification, Symmetry Analysis and Spatially Constrained Deformable Models. Fuzzy Set Syst 2009; 160: 1457-1473. DOI:10.1016/j.fss.2008.11.016
Dubey RB,Hanmandlu M,Gupta SK,Gupta SK. The Brain MR Image Segmentation Techniques and use of Diagnostic Packages. Acad Radiol. 2010; 17: 658 – 671. DOI:10.1016/j.acra.2009.12.017
Getreuer P. Linear Methods for Image Interpolation, Image Processing On Line, 2011; 1:238–259. DOI:10.5201/ipol.2011.g_lmii
Glassner AS. Graphics Gems. 1st ed., Academic Press Inc., 1993.
Dhal KG, Das A, Ray S, Galvez J, Das S. Nature Inspired Optimization Algorithms and Their Application in Multi Thresholding Image Segmentation. Archives of Computational Methods in Engineering (2020) 27:855–888. DOI: 10.1007/s11831-019-09334-y
Otsu N. A threshold selection method from gray-level histograms. IEEE T Syst Man Cyb 1979; 9: 62–66. DOI: 10.1109/TSMC.1979.4310076
Wagsta K, Cardie C, Rogers S, Schroedl S. .Constrained K-means Clustering with Background Knowledge. In: Proc. of the Eighteenth International Conference on Machine Learning (ICML 2001). 2001 ;1: 577-584.
Dempster AP, Laird NM, Rubin DB. Maximum Likelihood from Incomplete Data via the EM Algorithm. J R Stat Soc 1977; 39: 1–38. DOI: 10.1111/j.2517-6161.1977.tb01600.x
Wen Y, He L, Von Deneen KM, Lu Y. Brain tissue classification based on DTI using an improved Fuzzy C-means algorithm with spatial constraints. Magn Reson Imaging 2013; 31: 1623-1630. DOI: 10.1016/j.mri.2013.05.007
Demirkaya O, Asyali MK, Sahoo PK. Image Processing with MATLAB: Application in Medicine and Biology. USA: CRC Press, 2009.
Elaff I, El-Kemany A, Kholif M. Universal and stable medical image generation for tissue segmentation (The Unistable method). Turk J Electr Eng & Comp Sci 2017; 25: 1070-1081. DOI:10.3906/elk-1509-100.
Nguyen DMH, Vu HT, Ung HQ, Nguyen BT. 3D-brain segmentation using deep neural network and Gaussian mixture model. In Proc. IEEE Winter Conf. Appl. Comput. Vis. (WACV). 2017; 1: 815–824. DOI:10.1109/WACV.2017.96
Huo Y, Xu Z, Xiong Y, Aboud K, Parvathaneni P, Bao S, Bermudez C, Resnick SM, Cutting LE, Landman BA. 3D whole brain segmentation using spatially localized atlas network tiles. NeuroImage 2019; 194: 105-119. DOI: 10.1016/j.neuroimage.2019.03.041
Ramzan F, Khan MUG , Iqbal S, Saba T, Rehman A. Volumetric Segmentation of Brain Regions From MRI Scans Using 3D Convolutional Neural Networks. IEEE Access. 2020; 8:103697-103709. DOI:10.1109/ACCESS.2020.2998901
Kong Y, Chen X, Wu J, Zhang P, Chen Y, Shu H. Automatic brain tissue segmentation based on graph filter. BMC Med Imaging. 2018; 18(1):9. DOI:10.1186/s12880-018-0252-x
Kamarujjaman and Maitra M. 3D unsupervised modified spatial fuzzy c-means method for segmentation of 3D brain MR image. Pattern Anal Applic 2019; 22: 1561–1571. DOI:10.1007/s10044-019-00806-2
O’Donnell LJ, Westin CF. An Introduction to Diffusion Tensor Image Analysis. Neurosurgery Clinics of North America. 2011; 22(2):185-196. DOI: 10.1016/j.nec.2010.12.004
Kingsley PB. Introduction to diffusion tensor imaging mathematics: Part I. Tensors, rotations, and eigenvectors. Concept Magn. Reson. 2006; 28:101–122. DOI: 10.1002/cmr.a.20048
Basser PJ, Mattiello J, LeBihan D. MR Diffusion Tensor Spectroscopy and Imaging. Biophys J 1994; 66: 259-267. DOI: 10.1016/S0006-3495(94)80775-1
Basser PJ, Pierpaoli C. Microstructural and Physiological Features of Tissues Elucidated by Quantitative-Diffusion-Tensor MRI. J Magn Reson 1996; 111: 209–219. DOI: 10.1006/jmrb.1996.0086
Vilanova A, Zhang S, Kindlmann G, Laidlaw D. An Introduction to Visualization of Diffusion Tensor Imaging and Its Applications. In: Weickert, Joachim, Hagen, Hans. Visualization and Processing of Tensor Fields. Berlin Heidelberg: Springer, 2005. pp 121 – 153.
Holtmannspotter M, Peters N, Opherk C, Martin D, Herzog J, Bruckmann H, Samann P, Gschwendtner A, Dichgans M. Diffusion Magnetic Resonance Histograms as a Surrogate Marker and Predictor of Disease Progression in CADASIL: A Two-Year Follow-Up Study. Stroke 2005; 36: 2559-2565. DOI: 10.1161/01.STR.0000189696.70989.a4.
Seehaus A, Roebroeck A, Bastiani M, Fonseca L, Bratzke H, Lori N, Vilanova A, Goebel R, Galuske R. Histological validation of high-resolution DTI in human post mortem tissue. Front Neuroanat; 2015; 9:98. DOI: 10.3389/fnana.2015.00098
Dyrby TB, Lundell H, Burke MW, Reislev NL, Paulson OB, Ptito M, Siebner HR. Interpolation of diffusion weighted imaging datasets. Neuroimage, 2014: 103:202–213. DOI: 10.1016/j.neuroimage.2014.09.005

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Medical image enhancement based on volumetric tissue segmentation fusion (Uni-stable 3D method)

Status:

Version 1

Abstract

Figures

1. Introduction

2. Materials And Methods

2.1. Diffusion Tensor Imaging

2.2 3D Uni-stable Images

2.3. CSF/non-CSF 3D Uni-stable-Image generation

2.4. WM/non-WM 3D Uni-stable-Image generation

3. Results

3.2 Non-Uni-stable 3D images Segmentation

3.3 Uni-stable 3D images Segmentation

4. Conclusion And Discussion

Abbreviation

Declarations

References

Additional Declarations

Status:

Version 1