A Stacked Generalization of 3D Orthogonal Deep Learning Convolutional Neural Networks for Improved Detection of White Matter Hyperintensities in 3D FLAIR Images.

BACKGROUND AND PURPOSE: Accurate and reliable detection of white matter hyperintensities and their volume quantification can provide valuable clinical information to assess neurologic disease progression. In this work, a stacked generalization ensemble of orthogonal 3D convolutional neural networks, StackGen-Net, is explored for improving automated detection of white matter hyperintensities in 3D T2-FLAIR images. MATERIALS AND METHODS: Individual convolutional neural networks in StackGen-Net were trained on 2.5D patches from orthogonal reformatting of 3D-FLAIR (n = 21) to yield white matter hyperintensity posteriors. A meta convolutional neural network was trained to learn the functional mapping from orthogonal white matter hyperintensity posteriors to the final white matter hyperintensity prediction. The impact of training data and architecture choices on white matter hyperintensity segmentation performance was systematically evaluated on a test cohort (n = 9). The segmentation performance of StackGen-Net was compared with state-of-the-art convolutional neural network techniques on an independent test cohort from the Alzheimer's Disease Neuroimaging Initiative-3 (n = 20). RESULTS: StackGen-Net outperformed individual convolutional neural networks in the ensemble and their combination using averaging or majority voting. In a comparison with state-of-the-art white matter hyperintensity segmentation techniques, StackGen-Net achieved a significantly higher Dice score (0.76 [SD, 0.08], F1-lesion (0.74 [SD, 0.13]), and area under precision-recall curve (0.84 [SD, 0.09]), and the lowest absolute volume difference (13.3% [SD, 9.1%]). StackGen-Net performance in Dice scores (median = 0.74) did not significantly differ (P = .22) from interobserver (median = 0.73) variability between 2 experienced neuroradiologists. We found no significant difference (P = .15) in white matter hyperintensity lesion volumes from StackGen-Net predictions and ground truth annotations. CONCLUSIONS: A stacked generalization of convolutional neural networks, utilizing multiplanar lesion information using 2.5D spatial context, greatly improved the segmentation performance of StackGen-Net compared with traditional ensemble techniques and some state-of-the-art deep learning models for 3D-FLAIR.

Fully Automated Segmentation of Globes for Volume Quantification in CT Images of Orbits using Deep Learning.

BACKGROUND AND PURPOSE: Fast and accurate quantification of globe volumes in the event of an ocular trauma can provide clinicians with valuable diagnostic information. In this work, an automated workflow using a deep learning-based convolutional neural network is proposed for prediction of globe contours and their subsequent volume quantification in CT images of the orbits. MATERIALS AND METHODS: An automated workflow using a deep learning -based convolutional neural network is proposed for prediction of globe contours in CT images of the orbits. The network, 2D Modified Residual UNET (MRes-UNET2D), was trained on axial CT images from 80 subjects with no imaging or clinical findings of globe injuries. The predicted globe contours and volume estimates were compared with manual annotations by experienced observers on 2 different test cohorts. RESULTS: On the first test cohort (n = 18), the average Dice, precision, and recall scores were 0.95, 96%, and 95%, respectively. The average 95% Hausdorff distance was only 1.5 mm, with a 5.3% error in globe volume estimates. No statistically significant differences (P = .72) were observed in the median globe volume estimates from our model and the ground truth. On the second test cohort (n = 9) in which a neuroradiologist and 2 residents independently marked the globe contours, MRes-UNET2D (Dice = 0.95) approached human interobserver variability (Dice = 0.94). We also demonstrated the utility of inter-globe volume difference as a quantitative marker for trauma in 3 subjects with known globe injuries. CONCLUSIONS: We showed that with fast prediction times, we can reliably detect and quantify globe volumes in CT images of the orbits across a variety of acquisition parameters.

Clinical Utility of a Novel Ultrafast T2-Weighted Sequence for Spine Imaging.

BACKGROUND AND PURPOSE: TSE-based T2-weighted imaging of the spine has long scan times. This work proposes a fast imaging protocol using variable refocusing flip angles, optimized for blurring and specific absorption rate. MATERIALS AND METHODS: A variable refocusing flip angle echo-train was optimized for the spine to improve the point spread function and minimize the specific absorption rate, yielding images with improved spatial resolution and SNR compared with the constant flip angle sequence. Data were acquired from 51 patients (35 lumbar, 16 whole-spine) using conventional TSE and the proposed sequence, with a single-shot variant for whole-spine. Noninferiority analysis was performed to evaluate the efficiency of the proposed technique. RESULTS: The proposed multishot sequence resulted in a 2× shorter scan time with a >1.5× lower specific absorption rate. The variable flip angle sequence was noninferior to the conventional TSE (P < .025) for all image-quality and clinical criteria except signal-to-noise ratio for the lumbar spine protocol. However, mean image scores for the TSE-variable refocusing flip angle were ≥4.3 for all criteria, and concordance analysis showed high agreement (>90%) with the TSE, indicating clinical equivalence. The single-shot sequence resulted in 4× shorter whole-spine scans, and image scores were ≥4.4 for all criteria, attesting to its clinical utility. CONCLUSIONS: We present a fast T2-weighted spine protocol using variable refocusing flip angles, including a single-shot variant. The sequences have better point spread function behavior than their constant flip angle counterparts and, being faster, should be less sensitive to patient motion, often seen in the longer TSE scans.

Chemical-shift imaging utilizing the positional shifts along the readout gradient direction.

In this work, we describe a method that uses the linear phase acquired during the readout period due to chemical shift to generate individual magnetic resonance (MR) images of chemically shifted species. The method utilizes sets of Fourier (or k-space) data acquired with different directions of the readout gradient and a postprocessing algorithm to generate chemical shift images. The methodology is developed for both Cartesian data acquisition and for radial data acquisition. The method is presented here for two chemically shifted species but it can be extended to more species. In this work, we present the theory, show the results in phantoms and in human images, and discuss the artifacts and signal-to-noise ratio of the images obtained with the technique.

Reconstruction of MR images from data acquired on a general nonregular grid by pseudoinverse calculation.

A minimum-norm least-squares image-reconstruction method for the reconstruction of magnetic resonance images from non-Cartesian sampled data is proposed. The method is based on a general formalism for continuous-to-discrete mapping and pseudoinverse calculation. It does not involve any regridding or interpolation of the data and therefore the methodology differs fundamentally from existing regridding-based methods. Moreover, the method uses a continuous representation of objects in the image domain instead of a discretized representation. Simulations and experiments show the possibilities of the method in both radial and spiral imaging. Simulations revealed that minimum-norm least-squares image reconstruction can result in a drastic decrease of artifacts compared with regridding-based reconstruction. Besides, both in vivo and phantom experiments showed that minimum-norm least-squares image reconstruction leads to contrast improvement and increased signal-to-noise ratio compared with image reconstruction based on regridding. As an appendix, an analytical calculation of the raw data corresponding to the well-known Shepp and Logan software head phantom is presented.

High-resolution diffusion imaging with DIFRAD-FSE (diffusion-weighted radial acquisition of data with fast spin-echo) MRI.

A novel MRI method, DIFRAD-FSE (diffusion-weighted radial acquisition of data with fast spin-echo), is demonstrated that enables rapid, high-resolution multi-shot diffusion-weighted MRI without significant artifacts due to motion. Following a diffusion-weighting spin-echo preparation period, multiple radial lines of Fourier data are acquired using spin-echo refocusing. Images can be reconstructed from the radial data set using a magnitude-only filtered back-projection reconstruction algorithm that removes phase errors due to motion. Results from human brain imaging demonstrate the ability of DIFRAD-FSE to acquire multiple radial lines of Fourier data each TR period without significant artifacts due to relaxation and to produce high-resolution diffusion-weighted MRI images without significant artifacts from motion.

Plasmalemmal pH-gradients in drug-sensitive and drug-resistant MCF-7 human breast carcinoma xenografts measured by 31P magnetic resonance spectroscopy.

31p Magnetic resonance spectroscopy (MRS) was employed to investigate tumor pH in xenografts of drug-sensitive and drug-resistant MCF-7 human breast carcinoma cells. Measured extracellular pH values were found to be lower than the intracellular pH in all three tumor types investigated. The magnitude of this acid-outside plasmalemmal pH gradient increased with increasing tumor size in tumors of two drug-resistant variants of MCF-7 cells, but not in tumors of the parent (drug-sensitive) cells. The partitioning of weak-base or weak-acid drug molecules across the plasma membrane of a tumor cell is dependent upon the acid-dissociation constant (pKa) of the drug as well as the plasmalemmal pH gradient. A large acid-outside pH gradient, such as those seen in MCF-7 xenografts, can exert a protective effect on the cell from weak-base drugs such as anthracyclines and Vinca alkaloids, which have pKa values of 7.5 to 9.5. The possibility of enhancing the therapeutic efficacy of weak-base drugs by dietary or metabolic manipulation of the extracellular pH, in order to reduce or reverse the plasmalemmal pH gradient, deserves investigation.

Early increases in breast tumor xenograft water mobility in response to paclitaxel therapy detected by non-invasive diffusion magnetic resonance imaging.

An important goal in cancer chemotherapy is to sensitively and quantitatively monitor the response of individual patients' tumors to successful, or unsuccessful, therapy so that regimens can be altered iteratively. Currently, tumor response is monitored by frank changes in tumor morphology, yet these markers take long to manifest and are not quantitative. Recent studies suggest that the apparent diffusion coefficient of water (ADCw), measured noninvasively with magnetic resonance imaging, is sensitively and reliably increased in response to successful CTx. In the present study, we investigate the combination chemotherapy response of human breast cancer tumor xenografts sensitive or resistant to Paclitaxel by monitoring changes in the ADCw. Our results indicate that there is a clear, substantial, and early increase in the ADCw after successful therapy in drug sensitive tumors and that there is no change in the ADCw in p-glycoprotein-positive tumors, which are resistant to Paclitaxel. The mechanism underlying these changes is unknown yet is consistent with apoptotic cell shrinkage and a concomitant increase in the extracellular water fraction.

MRI and NMR spectroscopy of the lipids of atherosclerotic plaque in rabbits and humans.

The early stages of atherosclerosis are characterized by the deposition of cholesteryl esters and triglycerides into the arterial wall. In the excised human atherosclerotic plaque these lipids are in a liquid-like state at body temperature and observable via MRI and NMR spectroscopy. To assess the ability of MRI to quantitatively image the lipids of atherosclerotic plaque in vivo, we have investigated eight New Zealand White rabbits fed atherogenic diets (2 weight (wt)% cholesterol, 1 wt% cholesterol + 6 wt% peanut oil, and 1 wt% cholesterol + 6 wt% com oil). Postmortem examination indicated that all rabbits developed atherosclerosis in the aorta. Except for one animal, magnetic resonance angiography showed no noticeable obstruction in the aorta. MRI was carried out in an attempt to image atherosclerotic plaque lipids directly, but no signal was detected in vivo. However, a plaque lipid signal was observed from excised tissue using a small diameter RF coil. 1H NMR spectroscopy of the atherosclerotic plaque from excised aortas indicated that the major fraction of plaque lipids in rabbits is not in a liquid state at physiological temperature and are only marginally MRI-visible compared to human plaque lipid. The differences in the MRI characteristics of rabbit and human plaque are due to differences in the fatty acid profile of the cholesteryl esters, chiefly a decrease of linoleic acid in rabbit lesions.

Analysis and comparison of motion-correction techniques in diffusion-weighted imaging.

Motion continues to be a significant problem in MRI, producing image artifacts that can severely degrade image quality. In diffusion-weighted imaging (DWI), the problem is amplified by the presence of large gradient fields used to produce the diffusion weighting. Three correction methods applicable for correction of specific classes of motion are described and compared. The first is based on a generalised projection onto convex sets (GPOCS) postprocessing algorithm. The second technique uses the collection of navigator echoes to track phase errors. The third technique is based on a radial-scan data acquisition combined with a modified projection-reconstruction algorithm. Although each technique corrects well for translations, the radial-scan method proves to be more robust when more complex motions are present. A detailed description of the causes of MR data errors caused by rigid body motion is included as an appendix.

Magnetic resonance imaging of lipid deposits in human atheroma via a stimulated-echo diffusion-weighted technique.

NMR images of subintimal lipid deposits within the vessel walls of atherosclerotic human aortas were obtained at 37 and 27 degrees C at 4.7 T. A combination of a stimulated-echo and pulsed-field gradients was used for suppressing the mobile tissue water relative to the less mobile tissue lipids. At 27 degrees C there was also a substantial reduction of the subintimal lipid signal intensity, which is consistent with the characteristic phase transition of cholesteryl esters in human atheroma. These results represent the first direct detection of lipid deposits in nonprotruding atherosclerotic lesions with NMR imaging.