Deriving new soft tissue contrasts from conventional MR images using deep learning

Versatile soft tissue contrast in magnetic resonance imaging is a unique advantage of the imaging modality. However, the versatility is not fully exploited. In this study, we propose a deep learning-based strategy to derive more soft tissue contrasts from conventional MR images obtained in standard clinical MRI. Two types of experiments are performed. First, MR images corresponding to different pulse sequences are predicted from one or more images already acquired. As an example, we predict T_1ρ weighted knee image from T₂ weighted image and/or T₁ weighted image. Furthermore, we estimate images corresponding to alternative imaging parameter values. In a representative case, variable flip angle images are predicted from a single T₁ weighted image, whose accuracy is further validated in quantitative T₁ map subsequently derived. To accomplish these tasks, images are retrospectively collected from 56 subjects, and self-attention convolutional neural network models are trained using 1104 knee images from 46 subjects and tested using 240 images from 10 other subjects. High accuracy has been achieved in resultant qualitative images as well as quantitative T₁ maps. The proposed deep learning method can be broadly applied to obtain more versatile soft tissue contrasts without additional scans or used to normalize MR data that were inconsistently acquired for quantitative analysis.     

Quantitative T2 Imaging of Plant Tissues By Means Of Multi-Echo MRI Microscopy

A method for quantitative T₂ imaging is presented which covers the large range of T₂ values in plants (5 to 2000 ms) simultaneously. The transverse relaxation is characterized by phase-sensitive measurement of many echo images in a multi-echo magnetic resonance imaging sequence. Up to 1000 signal-containing echo images can be measured with an inter-echo time of 2.5 ms at 0.47 T. Separate images of water density and of T₂ are obtained. Results on test samples, on the cherry tomato and on the stem of giant hogweed are presented. The effects of field strength, spatial resolution and echo time on the observed T₂ values is discussed. The combination of a relatively low magnetic field strength, short echo time and medium pixel resolution results in excellent T₂ contrast and in images hardly affected by susceptibility artifacts. The characterization of transverse relaxation by multi-echo image acquisition opens a new route for studies of water balance in plants.     

A multi-scale residual network for accelerated radial MR parameter mapping

A deep learning MR parameter mapping framework which combines accelerated radial data acquisition with a multi-scale residual network (MS-ResNet) for image reconstruction is proposed. The proposed supervised learning strategy uses input image patches from multi-contrast images with radial undersampling artifacts and target image patches from artifact-free multi-contrast images. Subspace filtering is used during pre-processing to denoise input patches. For each anatomy and relaxation parameter, an individual network is trained. in vivo T₁ mapping results are obtained on brain and abdomen datasets and in vivo T₂ mapping results are obtained on brain and knee datasets. Quantitative results for the T₂ mapping of the knee show that MS-ResNet trained using either fully sampled or undersampled data outperforms conventional model-based compressed sensing methods. This is significant because obtaining fully sampled training data is not possible in many applications. in vivo brain and abdomen results for T₁ mapping and in vivo brain results for T₂ mapping demonstrate that MS-ResNet yields contrast-weighted images and parameter maps that are comparable to those achieved by model-based iterative methods while offering two orders of magnitude reduction in reconstruction times. The proposed approach enables recovery of high-quality contrast-weighted images and parameter maps from highly accelerated radial data acquisitions. The rapid image reconstructions enabled by the proposed approach makes it a good candidate for routine clinical use.     

Study of γ-irradiation crosslinking ofcis-1,4-polybutadiene by NMR microscopy

Crosslink properties of ψ-irradiatedcis-1,4-polybutadiene (PB) were studied by NMR microscopy. Spin density images of the irradiated bulk materials show that crosslinking at the ?CH= atom dominates and the segments between two crosslinks are still flexible enough to twist, forming a dense material.T ₁ andT ₂ weighted images of these bulk materials show that theT ₁/T ₂ ratio increases upon irradiation. This is an indication of the increase of stiffness upon irradiation.¹H spin density images of the irradiated PB swollen in C₆D₆ show that the dimension of the cavities formed by irradiation is greater than the space resolution of the experiment.     

Molecular Diffusion in NMR Microscopy

The signal-to-noise ratio and the T₂ contrast in ¹H NMR microscopy are strongly affected by self-diffusion effects. Here, we investigate the free diffusion of water within imaging gradients. As a result we obtain an apparent relaxation time T₂ which in NMR microscopy is at least one order of magnitude smaller than the true T₂ value of water in the object. This apparent T₂ relaxation is considerably reduced by improving spatial resolution. We conclude that quantitative true T₂ values cannot be calculated from series of images with increasing echo time. Furthermore, from the knowledge of the apparent T₂, an optimum short echo time can be found in order to maximize signal-to-noise ratio. Our theoretical findings are confirmed by phantom experiments at 11.75 T field strength.     

Estimating Spatial Resolution of In Vivo MR Images Using Spatial Modulation of Magnetization

Spatial Modulation of Magnetization is shown to provide a means of estimating perceived spatial resolution directly in vivo. On the first magnetic resonance system tested, resolution in conventional spin echo images was found to be stability limited in the phase encoding direction and voxel limited (via the Nyquist sampling theorem) in the frequency encoding direction both in vitro and in vivo. As the voxel size approaches half the stripe separation, fringes of resolved and unresolved stripes are formed across the image. This phenomenon is explained and described mathematically. On a second magnetic resonance scanner, resolution in the phase encoding direction of fast spin echo images with centrically ordered phase encoding is shown to be voxel limited in substances with long T₂, with poorer resolution in substances with short T₂. Resolution in fast spin echo images with linearly ordered phase encoding was shown to be voxel limited in the phase encoding direction.     

Solvent aided NMR microscopy of γ-irradiatedcis 1,4-polybutadiene

Cross-linked structure of γ-irradiatedcis 1,4-polybutadiene (PBi) was studied by NMR microscopy with the aid of deuterated benzene. After PBi was swollen (PBiS) in C₆D₆, the¹H NMR linewidth of the polymer was reduced from 4 to 2 kHz. The images show that the relaxation times,T ₁ andT ₂, of PBiS are longer than those of PBi. The simultaneous increase ofT ₁ andT ₂ implies that the molecular chains of PBi become more mobile upon swelling. The enhanced mobility of the molecular chains may provide a possibility of the increase in space resolution.     

High-definition,single-scan 2D MRI in inhomogeneous fields using spatial encoding methods

An approach has been recently introduced for acquiring two-dimensional (2D) nuclear magnetic resonance images in a single scan, based on the spatial encoding of the spin interactions. This article explores the potential of integrating this spatial encoding together with conventional temporal encoding principles, to produce 2D single-shot images with moderate field of views. The resulting “hybrid” imaging scheme is shown to be superior to traditional schemes in non-homogeneous magnetic field environments. An enhancement of previously discussed pulse sequences is also proposed, whereby distortions affecting the image along the spatially encoded axis are eliminated. This new variant is also characterized by a refocusing of T₂^* effects, leading to a restoration of high-definition images for regions which would otherwise be highly dephased and thus not visible. These single-scan 2D images are characterized by improved signal-to-noise ratios and a genuine T₂ contrast, albeit not free from inhomogeneity distortions. Simple postprocessing algorithms relying on inhomogeneity phase maps of the imaged object can successfully remove most of these residual distortions. Initial results suggest that this acquisition scheme has the potential to overcome strong field inhomogeneities acting over extended acquisition durations, exceeding 100 ms for a single-shot image.     

Accelerating quantitative MR imaging with the incorporation of B1 compensation using deep learning

Quantitative magnetic resonance imaging (MRI) attracts attention due to its support to quantitative image analysis and data driven medicine. However, the application of quantitative MRI is severely limited by the long data acquisition time required by repetitive image acquisition and measurement of field map. Inspired by recent development of artificial intelligence, we propose a deep learning strategy to accelerate the acquisition of quantitative MRI, where every quantitative T₁ map is derived from two highly undersampled variable-contrast images with radiofrequency field inhomogeneity automatically compensated. In a multi-step framework, variable-contrast images are first jointly reconstructed from incoherently undersampled images using convolutional neural networks; then T₁ map and B₁ map are predicted from reconstructed images employing deep learning. Thus, the acceleration includes undersampling in every input image, a reduction in the number of variable contrast images, as well as a waiver of B₁ map measurement. The strategy is validated in T₁ mapping of cartilage. Acquired with a consistent imaging protocol, 1224 image sets from 51 subjects are used for the training of the prediction models, and 288 image sets from 12 subjects are used for testing. High degree of acceleration is achieved with image fidelity well maintained. The proposed method can be broadly applied to quantify other tissue properties (e.g. T₂, T_1ρ) as well.     

Comparison of tumor histology to dynamic contrast enhanced magnetic resonance imaging-based physiological estimates

Purpose

The purpose of this study was to compare histologically determined cellularity and extracellular space to dynamic contrast-enhanced magnetic resonance imaging (DCE MRI)-based maps of a two-compartment model's parameters describing tumor contrast agent extravasation, specifically tumor extravascular extracellular space (EES) volume fraction (v_e), tumor plasma volume fraction (v_p) and volume-normalized contrast agent transfer rate between tumor plasma and interstitium (K_TRANS/V_T).

Materials and Methods

Obtained v_e, v_p and K_TRANS/V_T maps were estimated from gadolinium diethylenetriamine penta-acetic acid DCE T₁-weighted gradient-echo images at resolutions of 469, 938 and 2500 μm. These parameter maps were compared at each resolution to histologically determined tumor type, and the high-resolution 469-μm maps were compared with automated cell counting using Otsu's method and a color-thresholding method for estimated intracellular (V_{intracellular}) and extracellular (V_{extracellular}) space fractions.

Results

The top five K_TRANS/V_T values obtained from each tumor at 469 and 938 μm resolutions are significantly different from those obtained at 2500 μm (P<.0001) and from one another (P=.0014). Using these top five K_TRANS/V_T values and the corresponding tumor EES volume fractions v_e, we can statistically differentiate invasive ductal carcinomas from noninvasive papillary carcinomas for the 469- and 938-μm resolutions (P=.0017 and P=.0047, respectively), but not for the 2500-μm resolution (P=.9008). The color-thresholding method demonstrated that v_e measured by DCE MRI is statistically similar to histologically determined EES. The V_{extracellular} obtained from the color-thresholding method was statistically similar to the v_e measured with DCE MRI for the top 10 K_TRANS/V_T values (P>.05). DCE MRI-based K_TRANS/V_T estimates are not statistically correlated with histologically determined cellularity.

Conclusion

DCE MRI estimates of tumor physiology are a limited representation of tumor histological features. Extracellular spaces measured by both DCE MRI and microscopic analysis are statistically similar. Tumor typing by DCE MRI is spatial resolution dependent, as lower resolutions average out contributions to voxel-based estimates of K_TRANS/V_T. Thus, an appropriate resolution window is essential for DCE MRI tumor diagnosis. Within this resolution window, the top K_TRANS/V_T values with corresponding v_e are diagnostic for the tumor types analyzed in this study.     

Relaxation-Time and Diffusion NMR Microscopy of Single Neurons

Relaxation-time and diffusion-weighted NMR micrographs have been obtained for single neurons isolated from Aplysia californica. These images allow the nucleus and cytoplasm to be clearly differentiated, in contrast to proton spin-density images, which appear relatively homogenous. Images of the spatial distribution of T₁ and T₂ relaxivities and the diffusion coefficient (D), as well as average values for T₁, T₂, and D in the cytoplasm and nucleus, were calculated from sets of appropriately weighted images. In all cases, water in the nucleus had relaxation and diffusion properties markedly differing from those of cytoplasmic water, which in turn had properties which were distinct from those of free water. Additionally, the cytoplasmic T₂ was observed to triple following cell death, which is attributed to cytoplasmic dilution as water enters the cell. The work presented represents the first effort at a consistent exploration of the spatial distribution of NMR characteristics of water within intact single cells. These studies have implications both for modeling the NMR characteristics of water in neuronal tissues based on an understanding of the characteristics of water in different cell compartments and for understanding water/macromolecule interactions within cells. NMR microscopy studies such as these may help form a foundation for understanding and interpreting NMR characteristics measured from large assemblies of cells, i.e., spectroscopy and imaging of living tissues.     

Accelerated electron paramagnetic resonance imaging using partial Fourier compressed sensing reconstruction

PurposeElectron paramagnetic resonance (EPR) imaging has evolved as a promising tool to provide non-invasive assessment of tissue oxygenation levels. Due to the extremely short T₂ relaxation time of electrons, single point imaging (SPI) is used in EPRI, limiting achievable spatial and temporal resolution. This presents a problem when attempting to measure changes in hypoxic state. In order to capture oxygen variation in hypoxic tissues and localize cycling hypoxia regions, an accelerated EPRI imaging method with minimal loss of information is needed.MethodsWe present an image acceleration technique, partial Fourier compressed sensing (PFCS), that combines compressed sensing (CS) and partial Fourier reconstruction. PFCS augments the original CS equation using conjugate symmetry information for missing measurements. To further improve image quality in order to reconstruct low-resolution EPRI images, a projection onto convex sets (POCS)-based phase map and a spherical-sampling mask are used in the reconstruction process. The PFCS technique was used in phantoms and in vivo SCC7 tumor mice to evaluate image quality and accuracy in estimating O₂ concentration.ResultsIn both phantom and in vivo experiments, PFCS demonstrated the ability to reconstruct images more accurately with at least a 4-fold acceleration compared to traditional CS. Meanwhile, PFCS is able to better preserve the distinct spatial pattern in a phantom with a spatial resolution of 0.6 mm. On phantoms containing Oxo63 solution with different oxygen concentrations, PFCS reconstructed linewidth maps that were discriminative of different O₂ concentrations. Moreover, PFCS reconstruction of partially sampled data provided a better discrimination of hypoxic and oxygenated regions in a leg tumor compared to traditional CS reconstructed images.ConclusionsEPR images with an acceleration factor of four are feasible using PFCS with reasonable assessment of tissue oxygenation. The technique can greatly enhance EPR applications and improve our understanding cycling hypoxia. Moreover this technique can be easily extended to various MRI applications.     

Transverse relaxation time constraints on resolution in one-dimensional,phase- and frequency-encoded nuclear magnetic resonance imaging

The theoretical dependence of the resolution on the relationship of sampling time to transverse relaxation time (T₂) for frequency-encoded, one-dimensional NMR imaging using constant field gradients has been investigated. A resolution function that is explicitly dependent on the sampling time is derived, and it is shown that the observed image of an object can be written as a convolution of the sample magnetization with this resolution function. This function is explicitly calculated for two cases of interest: (1) for sampling times much shorter than T₂, and (2) for sampling times much longer than T₂. These cases are illustrated for two examples: (1) a uniform magnetic bar, and (2) uniform periodic magnetic bars. When oscillating gradients are utilized, these results still hold in the limit of slow oscillation. The resolution in phase-encoded NMR imaging is not explicitly dependent on the sampling time.     

Investigation of laminar appearance of articular cartilage by means of magnetic resonance microscopy

Magnetic resonance (MR) images and relaxation and diffusion maps of articular cartilage were obtained to explain discrepancies in its MR appearance. Porcine specimens were studied only by MR microscopy. For human specimens a combination of MR microscopy and large-scale MR imaging was used. Common features in the laminar structures of human and porcine samples are described. It was found that the decay of transverse magnetization was nonexponential with a rapidly decaying component which prevented construction of reliable proton-density maps. Dependence of T₂ values on the orientation of specimens in the magnetic field as well as magnetization transfer experiments supported the previous suggestions about a significant role of dipolar interaction with protons of collagen in the laminar appearance of articular cartilage. The loss of the laminar structure induced by rotation of the human cartilage specimen around the axis normal to its surface demonstrated nonuniform angular distribution of the collagen fibers within the layer.     

An investigation of the structural aspects of the tomato fruit by means of quantitative nuclear magnetic resonance imaging

In this study, magnetic resonance imaging (MRI) was applied to study the structural aspects of the tomato fruit. The main study was performed on tomatoes (cv. Tradiro) using a 0.2-T electromagnet scanner. Spin-echo images were acquired to visualize the tomato macrostructure. The air bubble content in tissues was evaluated by exploiting susceptibility effects using multiple gradient echo images. The microstructure was further studied by measuring spin–spin (T₂) and spin–lattice (T₁) relaxation time distributions. Nuclear magnetic resonance relaxometry, macro vision imaging and chemical analysis were used as complementary and independent experimental methods in order to emphasize the MRI results. MRI images showed that the air bubble content varied between tissues. The presence of gas was attested by macro vision images. Quantitative imaging showed that T₂ and T₁ maps obtained by MRI reflected the structural differences between tomato tissues and made it possible to distinguish between them. The results indicated that cell size and chemical composition contribute to the relaxation mechanism.     

Quantitative NMR spin-lattice-relaxation imaging of brine in sandstone reservoir cores

Two- and three-dimensional quantitative, saturation-recovery, NMR imaging has been applied to two sandstone reservoir cores. M₀ and T₁ images of high quality have been obtained with reasonable data-acquisition and data-processing times. The T₁ and T₂ processes have been shown to be correlated; the variations of the parameters within the images and the differences between the bulk and image relaxation values are discussed. The results of a quantitative T₁ imaging experiment of a phantom are also presented in order to demonstrate the advantages of quantitative NMR imaging over standard bulk measurements.     

High-performance rapid MR parameter mapping using model-based deep adversarial learning

PurposeTo develop and evaluate a deep adversarial learning-based image reconstruction approach for rapid and efficient MR parameter mapping.MethodsThe proposed method provides an image reconstruction framework by combining the end-to-end convolutional neural network (CNN) mapping, adversarial learning, and MR physical models. The CNN performs direct image-to-parameter mapping by transforming a series of undersampled images directly into MR parameter maps. Adversarial learning is used to improve image sharpness and enable better texture restoration during the image-to-parameter conversion. An additional pathway concerning the MR signal model is added between the estimated parameter maps and undersampled k-space data to ensure the data consistency during network training. The proposed framework was evaluated on T₂ mapping of the brain and the knee at an acceleration rate R = 8 and was compared with other state-of-the-art reconstruction methods. Global and regional quantitative assessments were performed to demonstrate the reconstruction performance of the proposed method.ResultsThe proposed adversarial learning approach achieved accurate T₂ mapping up to R = 8 in brain and knee joint image datasets. Compared to conventional reconstruction approaches that exploit image sparsity and low-rankness, the proposed method yielded lower errors and higher similarity to the reference and better image sharpness in the T₂ estimation. The quantitative metrics were normalized root mean square error of 3.6% for brain and 7.3% for knee, structural similarity index of 85.1% for brain and 83.2% for knee, and tenengrad measures of 9.2% for brain and 10.1% for the knee. The adversarial approach also achieved better performance for maintaining greater image texture and sharpness in comparison to the CNN approach without adversarial learning.ConclusionThe proposed framework by incorporating the efficient end-to-end CNN mapping, adversarial learning, and physical model enforced data consistency is a promising approach for rapid and efficient reconstruction of quantitative MR parameters.     

Planar sulfur-doped silicon detectors for high-speed infrared thermography

Planar extrinsic sulfur-doped silicon detectors for infrared (IR) semiconductor-discharge gap image converters intended for use in high-speed thermography of remote objects have been developed. The detectors were fabricated by high-temperature diffusion of sulfur into silicon wafers from the vapor phase. The dependence of doping efficiency on the sulfur vapor pressure in the course of diffusion was analyzed. The detector fabrication technology was optimized to meet the specific requirements for their operation in the microdischarge devices considered. The detectors were tested in a laboratory setup comprising a blackbody source of IR light, an image converter, and a pulsed CCD camera for recording the converted images. The converter equipped with the detector can provide imaging of objects heated to a temperature, T_min ≈ 200 °C, with a temporal resolution on the order of 10^?6 s and spatial resolution of about 5 lines/mm.     

Surface morphology of metal films deposited on mica at various temperatures observed by atomic force microscopy

Thin films of eight metals with a thickness of 150 nm were deposited on mica substrates by thermal evaporation at various temperatures in a high vacuum. The surface morphology of the metal films was observed by atomic force microscopy (AFM) and the images were characterized quantitatively by a roughness analysis and a bearing analysis (surface height analysis). The films of Au, Ag, Cu, and Al with the high melting points were prepared at homologous temperatures T/T_m = 0.22-0.32, 0.40, and 0.56. The films of In, Sn, Bi, and Pb with the low melting points were prepared at T/T_m = 0.55-0.70, where T and T_m are the absolute temperatures of the mica substrate and the melting point of the metal, respectively. The surface morphology of these metal films was studied based on a structure zone model. The film surfaces of Au, Ag, and Cu prepared at the low temperatures (T/T_m = 0.22-0.24) consist of small round grains with diameters of 30-60 nm and heights of 2-7 nm. The surface heights of these metal films distribute randomly around the surface height at 0 nm and the morphology is caused by self-shadowing during the deposition. The grain size becomes large due to surface diffusion of adatoms and the film surfaces have individual characteristic morphology and roughnesses as T increases. The surface of the Al film becomes very smooth as T increases and the atomically smooth surface is obtained at T/T_m = 0.56-0.67 (250-350 °C). On the other hand, the atomically smooth surface of the Au film is obtained at T/T_m = 0.56 (473 ± 3 °C). The films of In, Sn, Bi, and Pb prepared at T/T_m = 0.55-0.70 also show the individual characteristic surface morphology.     

Magnetic Resonance Imaging of Gases: A Single-Point Ramped Imaging with T1 Enhancement (SPRITE) Study

A pure phase-encoding MRI technique, single-point ramped imaging withT₁enhancement, SPRITE, is introduced for the purpose of gas phase imaging. The technique utilizes broadband RF pulses and stepped phase encode gradients to produce images, substantially free of artifacts, which are sensitive to the gasT₁andT^*:₂relaxation times. Images may be acquired from gas phase species with transverse relaxation times substantially less than 1 ms. Methane gas images,¹H, were acquired in a phantom study. Sulfur hexafluoride,¹⁹F, images were acquired from a gas-filled porous coral sample. High porosity regions of the coral are observed in both the MRI image and an X-ray image. Sensitivity and resolution effects due to signal modulation during the time-efficient acquisition are discussed. A method to increase the image sensitivity is discussed, and the predicted improvement is shown through 1D images of the methane gas phantom.