Automatic segmentation for brain MR images via a convex optimized segmentation and bias field correction coupled model

Accurate segmentation of magnetic resonance (MR) images remains challenging mainly due to the intensity inhomogeneity, which is also commonly known as bias field. Recently active contour models with geometric information constraint have been applied, however, most of them deal with the bias field by using a necessary pre-processing step before segmentation of MR data. This paper presents a novel automatic variational method, which can segment brain MR images meanwhile correcting the bias field when segmenting images with high intensity inhomogeneities. We first define a function for clustering the image pixels in a smaller neighborhood. The cluster centers in this objective function have a multiplicative factor that estimates the bias within the neighborhood. In order to reduce the effect of the noise, the local intensity variations are described by the Gaussian distributions with different means and variances. Then, the objective functions are integrated over the entire domain. In order to obtain the global optimal and make the results independent of the initialization of the algorithm, we reconstructed the energy function to be convex and calculated it by using the Split Bregman theory. A salient advantage of our method is that its result is independent of initialization, which allows robust and fully automated application. Our method is able to estimate the bias of quite general profiles, even in 7T MR images. Moreover, our model can also distinguish regions with similar intensity distribution with different variances. The proposed method has been rigorously validated with images acquired on variety of imaging modalities with promising results.     

自导航快速自旋回波在低场MRI上的实现

低场磁共振成像仪一般需采用数据累加的办法来提高图像信噪比,这样会延长扫描时间,因此更易受运动伪影的影响. 为了解决运动伪影问题,本文在低场磁共振成像仪上实现了自导航快速自旋回波去运动伪影成像技术,并且与常规快速自旋回波序列进行了临床对比实验. 结果表明,与常规快速自旋回波序列相比,采用自导航快速自旋回波技术后,由于病人运动导致的伪影得到明显地抑制.      

Ghost imaging with white light speckle patterns

In this paper we describe a method to study the formation of a ghost image by using a simplified setup. A spatial light modulator (SLM) working in amplitude regime is used both to display the object and the intensity distribution that provides the high-spatially correlated beams which are the base of this phenomenon. We compare the recovered ghost image with that obtained by using pseudothermal light. Moreover we show that good quality images are also retrieved when a turbulent media is present.     

Image-Based Reduction of Artifacts in Multishot Echo-Planar Imaging

The method to reduce the ghost artifact in echo-planar imaging (EPI) using a phase correction derived from the image data (M. H. Buonocore and L. Gao,Magn. Reson. Med.38, 89 (1997)) is generalized to multishot (interleaved) EPI, where the artifact takes the form of multiple ghosts. The method is shown to be much more sensitive to noise when applied to standard interleaved data than is the case with single-shot EPI, because the calculation must be based on high-order ghosts of low intensity. A modified interleaving scheme is proposed for multishot EPI in which the initial trajectory direction alternates in consecutive shots and the number of shots is odd. With this scheme, only a single ghost shifted by one-half of the field of view appears just as in the single-shot EPI, and the image-based phase correction can be applied with the usual sensitivity to noise.     

A least squares quantization table method for direct reconstruction of MR images with non-Cartesian trajectory

The direct Fourier transform method is a straightforward solution with high accuracy for reconstructing magnetic resonance (MR) images from nonuniformly sampled k-space data, given that the optimal density compensation function is selected and the underlying magnetic field is sufficiently uniform. The computation however is very time-consuming, making it impractical especially for large-size images. In this paper, the least squares quantization table (LSQT) method is proposed to accelerate the direct Fourier transform computation, similar to the recently proposed methods such as using look-up table (LUT) or equal-phase-line (EPL). With LSQT, all the image pixels are first classified into several groups where the Lloyd-Max quantization scheme is used to ensure the minimal classification error. The representative value of each group is stored in a small-size LSQT in advance to reduce the computational load. The pixels in the same group receive the same contribution, which is calculated only once for each group instead of for each pixel, resulting in the reduction of computation because the number of groups is far smaller than the number of pixels. Finally, each image pixel is mapped into the nearest group and its representative value is used to reconstruct the image. The experimental results show that the LSQT method requires far smaller memory size than the LUT method and fewer multiplication operations than the LUT and EPL methods. Moreover, the LSQT method can perform large-size reconstructions that achieve comparable or higher accuracy as compared to the EPL and gridding methods when the appropriate parameters are given. The inherent parallel structure also makes the LSQT method easily adaptable to a multiprocessor system.     

Magnetic resonance spectroscopic imaging for visualization and correction of distortions in MRI: high precision applications in neurosurgery

We present a method for the quantification and correction of geometrical/intensity distortions of magnetic resonance images predominantly caused by bulk magnetic susceptibility shifts due to susceptibility heterogeneities of measured biologic tissues and shape of the object under investigation. The method includes precise and fast measurements of the static magnetic-field distribution inside the measured object and automated data processing. Magnetic-field deviations in the range −2.4; 2.6 ppm were found in the human brain at B₀ = 1.5 T. For routinely used imaging parameters, with a read gradient strength of about 1 mT/m, the magnetic-field perturbations in the human brain can cause geometrical distortions up to ±4 mm and intensity changes up to ±50%. MR images corrected by the described method are suitable for planning high precision applications in neurosurgery.     

实验研究探测器大小与散焦长度对无透镜鬼衍射的影响

无透镜鬼衍射是一种特殊的非局域关联成像技术,通过对包含物体信息的测试光路和不含物体信息的参考光路的光场强度涨落关联进行测量,在参考光路上可得到物体的衍射图样。利用赝热光源实验研究了探测器尺寸大小和光路散焦长度对无透镜鬼衍射质量的影响。实验以四缝物体为例,定量研究了探测器大小和散焦是如何改变无透镜鬼衍射的成像结果,并和理论模拟进行了比较,获得了比较一致的结果。利用图像相关度定量分析了实验结果与理想无透镜鬼衍射图像的差别,发现这两者都会使无透镜鬼衍射的质量变差。     

Effects of light intensity on reflective ghost imaging

In this letter, we analyze the effects of light intensity find that the brightness of reflective ghost image can on reflective ghost imaging with thermal source. We be changed by modulating the light intensity of the source and the splitting ratio of the beam splitter. The signal-to-noise ratio will be improved by increa.sing the light intensity of the source. More important, we can obtain the reflective ghost image with high image quality by adopting a low light intensity signal beam and a high light intensity reference beam, which is better than the classical optical imaging, because it can reduce the effects of light on the object.     

A novel, fast entropy-minimization algorithm for bias field correction in MR images

A novel, fast entropy-minimization algorithm for bias field correction in magnetic resonance (MR) images is suggested to correct the intensity inhomogeneity degradation of MR images that has become an increasing problem with the use of phased-array coils. Four important modifications were made to the conventional algorithm: (a) implementation of a modified two-step sampling strategy for stacked 2D image data sets, which included reducing the size of the measured image on each slice with a simple averaging method without changing the number of slices and then using a binary mask generated by a histogram threshold method to define the sampled voxels in the reduced image; (b) improvement of the efficiency of the correction function by using a Legendre polynomial as an orthogonal base function polynomial; (c) use of a nonparametric Parzen window estimator with a Gaussian kernel to calculate the probability density function and Shannon entropy directly from the image data; and (d) performing entropy minimization with a conjugate gradient method. Results showed that this algorithm could correct different types of MR images from different types of coils acquired at different field strengths very efficiently and with decreased computational load.     

Correlations of Multiplexed Quantum Ghost Images and Improvement of the Quality of Restored Image

The currently used ghost-image schemes traditionally involve two-mode entangled light states or incoherent radiation. Here, we consider the application of four-mode entangled light states and show that multiplexed ghost images (MGI) formed by four-mode entangled quantum light states have mutual spatial correlations determined by the eighth-order field correlation functions. We develop a special algorithm to calculate high-order correlations of Bose operators. We also demonstrate that accounting of the MGI correlations allows us to improve the quality of the restored image of an object while processing the MGI by the measurement reduction method. We carry out computer modeling of the image recovery from the MGI. We establish that in the considered example the signal-to-noise ratio of the reduced ghost image is 4.6 times higher than the best signal-to-noise ratio for the ghost images themselves.     

结合小波融合和深度学习的脑胶质瘤自动分割

针对水肿区域边界模糊和瘤内结构复杂多变导致的脑胶质瘤分割不精确问题,本文提出了一种基于小波融合和3D-UNet网络的脑胶质瘤磁共振图像自动分割算法.首先,对脑胶质瘤磁共振图像的T1、T1ce、T2、Flair四种模态进行小波融合以及偏置场校正;然后,提取待分类的图像块;再利用提取的图像块训练3D-UNet网络以对图像块中的像素进行分类;最后加载损失率较小的网络模型进行分割,并采用基于连通区域的轮廓提取方法,以降低假阳性率.对57组Brats2018（Brain Tumor Segmentation 2018）磁共振图像测试集进行分割的结果显示,肿瘤的整体、核心和水肿部分的平均分割准确率（DSC）分别达到90.64%、80.74%和86.37%,这表明该算法分割脑胶质瘤准确率较高,与金标准相近.相比多模态图像融合前,该算法在减少输入网络数据量和图像冗余信息的同时,还一定程度上解决了胶质瘤边界模糊、分割不精确的问题,提高了分割的准确度和鲁棒性.     

Multiplicative intrinsic component optimization (MICO) for MRI bias field estimation and tissue segmentation

This paper proposes a new energy minimization method called multiplicative intrinsic component optimization (MICO) for joint bias field estimation and segmentation of magnetic resonance (MR) images. The proposed method takes full advantage of the decomposition of MR images into two multiplicative components, namely, the true image that characterizes a physical property of the tissues and the bias field that accounts for the intensity inhomogeneity, and their respective spatial properties. Bias field estimation and tissue segmentation are simultaneously achieved by an energy minimization process aimed to optimize the estimates of the two multiplicative components of an MR image. The bias field is iteratively optimized by using efficient matrix computations, which are verified to be numerically stable by matrix analysis. More importantly, the energy in our formulation is convex in each of its variables, which leads to the robustness of the proposed energy minimization algorithm. The MICO formulation can be naturally extended to 3D/4D tissue segmentation with spatial/sptatiotemporal regularization. Quantitative evaluations and comparisons with some popular softwares have demonstrated superior performance of MICO in terms of robustness and accuracy.     

用于压缩感知磁共振成像的分割字典学习算法

字典学习算法可以根据数据本身的特点构建稀疏域中的基,从而使数据的表示更加稀疏.该文在传统的字典学习算法基础上提出了分割字典学习算法,由于部分磁共振图像组织结构简单、可以进行图像分割,因此可根据此特点来优化字典中基函数的构建,使磁共振图像的表达更为稀疏,从而获得更高的重建图像质量.该文利用模拟数据和真实数据进行了重建实验,结果表明与传统的字典学习算法相比,分割字典学习算法能进一步改善重建图像质量.     

Incorporating prior knowledge via volumetric deep residual network to optimize the reconstruction of sparsely sampled MRI

For sparse sampling that accelerates magnetic resonance (MR) image acquisition, non-linear reconstruction algorithms have been developed, which incorporated patient specific a prior information. More generic a prior information could be acquired via deep learning and utilized for image reconstruction. In this study, we developed a volumetric hierarchical deep residual convolutional neural network, referred to as T-Net, to provide a data-driven end-to-end mapping from sparsely sampled MR images to fully sampled MR images, where cartilage MR images were acquired using an Ultra-short TE sequence and retrospectively undersampled using pseudo-random Cartesian and radial acquisition schemes. The network had a hierarchical architecture that promoted the sparsity of feature maps and increased the receptive field, which were valuable for signal synthesis and artifact suppression. Relatively dense local connections and global shortcuts were established to facilitate residual learning and compensate for details lost in hierarchical processing. Additionally, volumetric processing was adopted to fully exploit spatial continuity in three-dimensional space. Data consistency was further enforced. The network was trained with 336 three-dimensional images (each consisting of 32 slices) and tested by 24 images. The incorporation of a priori information acquired via deep learning facilitated high acceleration factors (as high as 8) while maintaining high image fidelity (quantitatively evaluated using the structural similarity index measurement). The proposed T-Net had an improved performance as compared to several state-of-the-art networks.     

High-resolution imaging using Hadamard encoding

High-resolution imaging techniques using noninvasive modalities such as magnetic resonance (MR) imaging are being pursued as in vivo cancer screening techniques in an attempt to eliminate the invasive nature of surgical biopsy. When acquiring high-resolution MR images for tissue screening, image fields of view have in the past been limited by the matrix sizes available in conventional MR scanners. We present here a technique that uses aliasing to produce high resolution images with larger matrix sizes than are currently available. The image is allowed to alias in both the frequency encoding and phase encoding dimensions, and the individual, aliased fields of view are recovered by Hadamard encoding methods. These fields may then be tiled to obtain a composite image with high spatial resolution and a large field of view. The technique is demonstrated using two-dimensional and three-dimensional in vivo imaging of the human brain and breast.     

Referenceless PRFS MR Thermometry Using Partial Separability Model

Many areas of magnetic resonance (MR)-guided thermal therapy research would benefit from temperature maps with high spatial and temporal resolution. Conventional thermometry relies on the subtraction of baseline images, which makes it sensitive to tissue motion and frequency drift during the course of treatment. For another case is the limit of magnetic resonance imaging sampling speed, it is hard to accurately achieve MR thermometry with high spatiotemporal resolution especially for dynamic organs. To address these issues, a novel method for MR thermometry is presented by exploiting the data redundancy based on partial separability (PS) model and the referenceless thermometry. The PS model highly sparse sample two datasets in the (k–t) space for image reconstruction, which respectively determine the spatial and temporal resolutions. After the phase information is extracted from the images reconstructed by the PS model, the background phase outside the heated region from each acquired phase image through a polynomial fitting is estimated. Extrapolation of the polynomial to the heated region serves as the background phase estimate, which is then subtracted from the actual phase. The thermometry results showed that this method could accurately capture the dynamic change of MR thermometric images with 1.5 mm × 1.5 mm spatial resolution and 250 ms temporal resolution, respectively. The in vivo experiment of MR-guided high intensity focused ultrasound research and the cardiac dynamic MR thermometry are shown to demonstrate the benefits of the proposed method in high spatiotemporal resolution MR thermometry.     

Handcrafted fuzzy rules for tissue classification

This article proposes a handcrafted fuzzy rule-based system for segmentation and identification of different tissue types in magnetic resonance (MR) brain images. The proposed fuzzy system uses a combination of histogram and spatial neighborhood-based features. The intensity variation from one type of tissue to another is gradual at the boundaries due to the inherent nature of the MR signal (MR physics). A fuzzy rule-based approach is expected to better handle these variations and variability in features corresponding to different types of tissues. The proposed segmentation is tested to classify the pixels of the T2-weighted axial MR images of the brain into three primary tissue types: white matter, gray matter and cerebral-spinal fluid. The results are compared with those from manual segmentation by an expert, demonstrating good agreement between them.     

Synthetic T1-weighted brain image generation with incorporated coil intensity correction using DESPOT1

The increased use of phased-array and surface coils in magnetic resonance imaging, the push toward increased field strength and the need for standardized imaging across multiple sites during clinical trials have resulted in the need for methods that can ensure consistency of intensity both within the image and across multiple subjects/sites. Here, we describe a means of addressing these concerns through an extension of the rapid T(1) mapping technique - driven equilibrium single-pulse observation of T(1). The effectiveness of the proposed approach was evaluated using human brain T(1) maps acquired at 1.5 T with a multichannel phased-array coil. Corrected "synthetic" T(1)-weighted images were reconstructed by substituting the T(1) values back into the governing signal intensity equation while assuming a constant value for the equilibrium magnetization. To demonstrate signal normalization across a longitudinal study, we calculated synthetic T(1)-weighted images from data acquired from the same healthy subject at four different time points. Signal intensity profiles between the acquired and synthetic images were compared to determine the improvements with our proposed approach. Following correction, the images demonstrate obvious qualitative improvement with increased signal uniformity across the image. Near-perfect signal normalization was also observed across the longitudinal study, allowing direct comparison between the images. In addition, we observe an increase in contrast-to-noise ratio (compared with regular T(1)-weighted images) for synthetic images created, assuming uniform proton density throughout the volume. The proposed approach permits rapid correction for signal intensity inhomogeneity without significantly lengthening exam time or reducing image signal-to-noise ratio. This technique also provides a robust method for signal normalization, which is useful in multicenter longitudinal MR studies of disease progression, and allows the user to reconstruct T(1)-weighted images with arbitrary T(1) weighting.     

Wavelet-based characterization of vertebral trabecular bone structure from magnetic resonance images at 3 T compared with micro-computed tomographic measurements

Trabecular bone structure and bone density contribute to the strength of bone and are important in the study of osteoporosis. Wavelets are a powerful tool in characterizing and quantifying texture in an image. The purpose of this study was to validate wavelets as a tool in computing trabecular bone thickness directly from gray-level images. To this end, eight cylindrical cores of vertebral trabecular bone were imaged using 3-T magnetic resonance imaging (MRI) and micro-computed tomography (microCT). Thickness measurements of the trabecular bone from the wavelet-based analysis were compared with standard 2D structural parameters analogous to bone histomorphometry (MR images) and direct 3D distance transformation methods (microCT images). Additionally, bone volume fraction was determined using each method. The average difference in trabecular thickness between the wavelet and standard methods was less than the size of 1 pixel size for both MRI and microCT analysis. A correlation (R) of .94 for microCT measurements and that of .52 for MRI were found for the bone volume fraction. Based on these results, we conclude that wavelet-based methods deliver results comparable with those from established MR histomorphometric measurements. Because the wavelet transform is more robust with respect to image noise and operates directly on gray-level images, it could be a powerful tool for computing structural bone parameters from MR images acquired using high resolution and thus limited signal scenarios.