Unbiased segmentation of diffusion-weighted magnetic resonance images of the brain using iterative clustering

Segmentation of diffusion-weighted echo-planar imaging (DW-EPI) is challenging because of concerns regarding spatial resolution and distortion. Methods commonly used require manual input and often need thresholding measures to segment white matter (WM), gray matter (GM) and cerebrospinal fluid (CSF). This may introduce operator bias and misclassification error. When comparing patients with a diffuse disease process-such as multiple sclerosis (MS)--with healthy controls, although information from all images may be biased due to disease effect, this is more so if the data set employed to perform segmentation is also used as a measured outcome for the study, for example, fractional anisotropy maps. Presented in this work is an unbiased method for segmenting DW-EPI data sets using the b=0 and single-shot inversion recovery EPI into WM, GM and CSF. The method employs an iterative clustering technique to account for partial volume effects and signal variation caused by radiofrequency inhomogeneity. The technique is evaluated with both real and synthetic brain data and results compared with statistical parametric mapping (SPM02). With synthetic brain data, where a gold standard of segmentation exists, the presented method showed less misclassification compared to SPM02. The unbiased method proposed may provide a more accurate methodology of segmentation in the analysis of DWI-EPI images in conditions such as MS.     

Semiautomated quality assurance for quantitative magnetic resonance imaging.

It is now well established that MRI can be used for quantitative (as opposed to simply qualitative) measurements, and good accuracy and precision have been obtained in phantom experiments. To make routine quantitative measurements as part of a clinical scanning protocol, however, quality assurance (QA) methods particularly suited to quantification must be developed. We describe a set of QA tests using clinical protocols on test phantoms, with which we have assessed quantitative performance of our Picker 0.5-T scanner (Picker International, Cleveland, OH) over 2 years. We also describe the automated data processing methods we have developed to deal with the large amounts of data generated by these tests.     

A simple correction for B1 field errors in magnetization transfer ratio measurements

B1 errors are a problem in magnetization transfer ratio (MTR) measurements because the MTR value is dependent on the amplitude of the magnetization transfer (MT) pulse. B1 errors can arise from radiofrequency (RF) nonuniformity (caused by the RF coil, or skin effect and dielectric resonance in the subject's head) and also from inaccurate setting of the transmitter output when compensating for varying amounts of loading of the RF coil. B1 errors, and hence MTR errors, may be up to 5-10%, a large source of error in quantitative MR measurements. Radiofrequency nonuniformity may cause MTR histograms to be broadened. The dependence of MTR on B1 was modeled using binary spin bath theory, with a continuous wave (CW) approximation. For B1 reductions of up to 20%, normalized plots for different brain tissue types could be approximated by a single line, indicating that a systematic correction could be applied to MTR measurements with a known B1 error, regardless of tissue type. On a 1.5-T scanner with a birdcage coil, MTR was measured in 18 tissue types in five controls. The MT pulse amplitude was reduced in steps from its nominal value by up to 20%. Averaging data over all controls and tissue types resulted in a line fitting mtr(normalized)=0.812b(1normalized)+0.193, where mtr(normalized) is the normalized value of MTR (relative to its value at the nominal B1) and b(1normalized) is the normalized value of B1 (relative to its nominal value). For a 20% reduction in MT pulse amplitude (i.e., b(1normalized)=0.80), the mean MTR value for the 18 tissue types was 7.0 percent units (pu) below the correct value. After correction using the single equation above for all tissue types, all MTR values were within 1.5 pu of their correct value [root mean square (rms) error=0.7 pu]. Magnetization transfer ratio values tended to be slightly overcorrected because the simple linear correction scheme is only an approximation to the true MTR dependence on B1. A B1 field mapping technique was implemented, based on the double angle method (DAM), with fast spin-echo (FSE) readout, and TR=15 s; this took a total of 6 min of imaging time. This was used to quantify B(1) errors and correct MTR maps and histograms. However, the cerebrospinal fluid (CSF) T1 is very long (approximately 4.2 s); thus, to achieve complete longitudinal relaxation (a requirement of the DAM B1 mapping method), an increase in TR and, hence, acquisition time would be required. In general, however, we are not interested in calculating the B1 in the CSF, although it is important that the B1 is determined in partial volume voxels around the CSF. Using our birdcage head coil, whole-brain B1 histograms were found to have full-width half maximums (FWHMs) ranging from just 6.8% to 11.5% of the nominal B1 value. The FSE DAM B1 field mapping technique was shown to be robust, although a longer TR time may be desirable to ensure complete elimination of CSF partial volume errors. The procedure can be applied on any scanner where the Euro-MT sequence is available, or alternatively, where the amplitude of B1 or of the MT pulse can be manually reduced in order to perform this type of "calibration" experiment for the particular MTR sequence used. The MTR is known to be highly dependent on the parameters of the sequence used, in particular, the MT pulse shape, flip angle, duration, and offset frequency, and the repetition time TR' between successive MT pulses. Therefore, correction schemes will differ for different MTR sequences, and new data sets would be required to calculate these different correction schemes.     

Novel MR image contrast mechanisms in epilepsy

A range of magnetic resonance (MR) parameters are introduced, which can give rise to image contrast by using suitable pulse sequences, and that can be measured quantitatively. Their relationship to tissue pathology is given as far as possible. Techniques for their measurement, and results from multiple sclerosis, stroke, and epilepsy are given. The parameters are proton density, T₁, T₂, transverse magnetisation decay, which gives estimates of extracellular water and myelin concentrations, magnetisation transfer ratio and T_1sat, and diffusion (including trace and anisotropy measured from the tensor matrix).     

Precise estimate of fundamental in-vivo MT parameters in human brain in clinically feasible times

A methodology is presented for extracting precise quantitative MT parameters using a magnetisation-prepared spoiled gradient echo sequence. This method, based on a new mathematical model, provides relaxation parameters for human brain in-vitro and in-vivo. The in-vivo parameters have been obtained from three different regions of normal white matter: occipital white matter, frontal white matter and centrum semiovale; two regions of normal grey matter: cerebral cortex and cerebellum, and from five regions with MS lesions. All this has been achieved using MT images collected within a timeframe that is clinically feasible. We hope that this new technique will shed light on the properties and dynamics of water compartments within the brain.     

Sources of T1 variance in normal human white matter

The major factors contributing to T1 variance at 0.5 T in white matter were studied in healthy people. Anatomical location of the white matter sampled and differences between individuals contributed 74% of the total variance in serial measurements of the same subjects. There was also significant change over time within an individual subject that could not be attributed to machine drift. This information permitted estimates to be made concerning adequate sample size in future studies that examine for pathological white matter T1 change.     

An Extremal Property of Turán Graphs,II

Let denote Turán's graph—the complete 2‐partite graph on n vertices with partition sizes as equal as possible. We show that for all , the graph has more proper vertex colorings in at most 4 colors than any other graph with the same number of vertices and edges.     

The Bland–Altman analysis: Does it have a role in assessing radiation dosimeter performance relative to an established standard?

Bland–Altman analysis is used to compare two different methods of measurement and to determine whether a new method of measurement may replace an existing accepted ‘gold standard’ method. In this work, Bland–Altman analysis has been applied to radiation dosimetry to compare the PTW Markus and Roos parallel plate ionisation chambers and a PTW PinPoint chamber against a Farmer type ionisation chamber which is accepted as the gold standard for radiation dosimetry in the clinic. Depth doses for low energy x-rays beams with energies of 50, 75 and 100 kVp were measured using each of the ionisation chambers. Depth doses were also calculated by interpolation of the data in the British Journal of Radiology (BJR) Report 25. From the Bland–Altman analysis, the mean dose difference between the two parallel plate chambers and the Farmer chambers was 1% over the range of depths measured. The PinPoint chamber gave significant dose differences compared to the Farmer chamber. There were also differences of up to 12% between the BJR Report 25 depth doses and the measured data. For the Bland–Altman plots, the lines representing the limits of agreement were selected to be a particular percentage agreement e.g. 1 or 2%, instead of being based on the standard deviation (σ) of the differences. The Bland–Altman statistical analysis is a powerful tool for making comparisons of ionisation chambers with an ionisation chamber that has been accepted as a ‘gold standard’. Therefore we conclude that Bland–Altman analysis does have a role in assessing radiation dosimeter performance relative to an established standard.