Purpose: To make the quantitative blood oxygenation level-dependent (qBOLD) method more suitable for clinical application by accounting for proton diffusion and reducing acquisition times. Materials and Methods:Monte Carlo methods are used to simulate the signal from diffusing protons in the presence of a blood vessel network. A diffusive qBOLD model was then constructed using a lookup table of the results. Acquisition times are reduced by parallel imaging and by employing an integrated fieldmapping method, rather than running an additional sequence.Results: The addition of diffusion to the model is shown to have a significant impact on predicted signal formation. This is found to affect all fitted parameters when the model is applied to real data. Parallel imaging and integrated fieldmapping allowed the GESSE (gradient echo sampling of a spin echo) acquisition to be made in less than 10 minutes while maintaining high signal-to-noise ratio (SNR). Conclusion:By incorporating integrated field mapping and parallel imaging techniques, GESSE data were acquired within clinically acceptable acquisition times. These data fit closely to the diffusive qBOLD model, providing more realistic and robust measurements of T 2 and blood oxygenation than the static model.

Different theoretical models of the BOLD contrast mechanism are used for many applications including BOLD quantification (qBOLD) and vessel size imaging, both in health and disease. Each model simplifies the system under consideration, making approximations about the structure of the blood vessel network and diffusion of water molecules through inhomogeneities in the magnetic field created by deoxyhemoglobin-containing blood vessels. In this study, Monte-Carlo methods are used to simulate the BOLD MR signal generated by diffusing water molecules in the presence of long, cylindrical blood vessels. Using these simulations we introduce a new, phenomenological model that is far more accurate over a range of blood oxygenation levels and blood vessel radii than existing models. This model could be used to extract physiological parameters of the blood vessel network from experimental data in BOLD-based experiments. We use our model to establish ranges of validity for the existing analytical models of Yablonskiy and Haacke, Kiselev and Posse, Sukstanskii and Yablonskiy (extended to the case of arbitrary time in the spin echo sequence) and Bauer et al. (extended to the case of randomly oriented cylinders). Although these models are shown to be accurate in the limits of diffusion under which they were derived, none of them is accurate for the whole physiological range of blood vessels radii and blood oxygenation levels. We also show the extent of systematic errors that are introduced due to the approximations of these models when used for BOLD signal quantification.

The network activated during normal route learning shares considerable homology with the network of degeneration in the earliest symptomatic stages of Alzheimer's disease (AD). This inspired the virtual route learning test (VRLT) in which patients learn routes in a virtual reality environment. This study investigated the neural basis of VRLT performance in AD to test whether impairment was underpinned by a network or by the widely held explanation of hippocampal degeneration. VRLT score in a mild AD cohort was regressed against gray matter (GM) density and diffusion tensor metrics of white matter (WM) (n = 30), and, cerebral glucose metabolism (n = 26), using a mass univariate approach. GM density and cerebral metabolism were then submitted to a multivariate analysis [support vector regression (SVR)] to examine whether there was a network associated with task performance. Univariate analyses of GM density, metabolism and WM axial diffusion converged on the vicinity of the retrosplenial/posterior cingulate cortex, isthmus and, possibly, hippocampal tail. The multivariate analysis revealed a significant, right hemisphere-predominant, network level correlation with cerebral metabolism; this comprised areas common to both activation in normal route learning and early degeneration in AD (retrosplenial and lateral parietal cortices). It also identified right medio-dorsal thalamus (part of the limbic-diencephalic hypometabolic network of early AD) and right caudate nucleus (activated during normal route learning). These results offer strong evidence that topographical memory impairment in AD relates to damage across a network, in turn offering complimentary lesion evidence to previous studies in healthy volunteers for the neural basis of topographical memory. The results also emphasize that structures beyond the mesial temporal lobe (MTL) contribute to memory impairment in AD—it is too simplistic to view memory impairment in AD as a synonym for hippocampal degeneration.

This paper investigates the scaling properties of Recurrent Neural Network Language Models (RNNLMs). We discuss how to train very large RNNs on GPUs and address the questions of how RNNLMs scale with respect to model size, training-set size, computational costs and memory. Our analysis shows that despite being more costly to train, RNNLMs obtain much lower perplexities on standard benchmarks than n-gram models. We train the largest known RNNs and present relative word error rates gains of 18% on an ASR task. We also present the new lowest perplexities on the recently released billion word language modelling benchmark, 1 BLEU point gain on machine translation and a 17% relative hit rate gain in word prediction.

We present a technique for predicting cardiac and respiratory phase on a time point by time point basis, from fMRI image data. These predictions have utility in attempts to detrend effects of the physiological cycles from fMRI image data. We demonstrate the technique both in the case where it can be trained on a subject's own data, and when it cannot. The prediction scheme uses a multiclass support vector machine algorithm. Predictions are demonstrated to have a close fit to recorded physiological phase, with median Pearson correlation scores between recorded and predicted values of 0.99 for the best case scenario (cardiac cycle trained on a subject's own data) down to 0.83 for the worst case scenario (respiratory predictions trained on group data), as compared to random chance correlation score of 0.70. When predictions were used with RETROICOR--a popular physiological noise removal tool--the effects are compared to using recorded phase values. Using Fourier transforms and seed based correlation analysis, RETROICOR is shown to produce similar effects whether recorded physiological phase values are used, or they are predicted using this technique. This was seen by similar levels of noise reduction noise in the same regions of the Fourier spectra, and changes in seed based correlation scores in similar regions of the brain. This technique has a use in situations where data from direct monitoring of the cardiac and respiratory cycles are incomplete or absent, but researchers still wish to reduce this source of noise in the image data.