Imaging biomarkers now play a key role in both basic and applied research and are increasingly seen as important in diagnosis, drug discovery and therapeutic trials. In collaboration with the Dementia Research Centre (N. Fox, M. Rossor, J. Schott) we are developing novel imaging biomarkers for Alzheimer’s disease using MR and PET images, enabling the robust and automatic measure of disease progression and disease modification.
We have been focusing on whole brain atrophy , hippocampus, cortical thickness, longitudnal analysis, DTI and more recently PIB uptake, fMRI analysis and classification. These imaging biomarkers are currently used in many clinical trials, including large international activities, such as the ADNI and DIAN initiatives.
GLOBAL AND LOCAL ATROPHY
Brain atrophy measurement is increasingly important in studies of neurodegenerative diseases such as Alzheimer's disease (AD), with particular relevance to trials of potential disease-modifying drugs. Automated registration-based methods such as the boundary shift integral (BSI) have been developed to provide more precise measures of change from a pair of serial MR scans. However, when a method treats one image of the pair (typically the baseline) as the reference to which the other is compared, this systematic asymmetry risks introducing bias into the measurement. Recent concern about potential biases in longitudinal studies has led to several suggestions to use symmetric image registration, though some of these methods are limited to two time-points per subject. Therapeutic trials and natural history studies increasingly involve several se- rial scans, it would therefore be useful to have a method that can consistently estimate brain atrophy over multiple time-points. Here, we use the log-Euclidean concept of a within-subject average to develop affine registration and differential bias correction methods suitable for any number of time-points, yielding a lon- gitudinally consistent multi-time-point BSI technique. Baseline, 12-month and 24-month MR scans of healthy controls, subjects with mild cognitive impairment and AD patients from the Alzheimer's Disease Neuroimaging Initiative are used for testing the bias in processing scans with different amounts of atrophy. Four tests are used to assess bias in brain volume loss from BSI: (a) inverse consistency with respect to ordering of pairs of scans 12 months apart; (b) transitivity consistency over three time-points; (c) randomly ordered back-to- back scans, expected to show no consistent change over subjects; and (d) linear regression of the atrophy rates calculated from the baseline and 12-month scans and the baseline and 24-month scans, where any additive bias should be indicated by a non-zero intercept. Results indicate that the traditional BSI processing pipeline does not exhibit significant bias due to its use of windowed sinc interpolation, but with linear interpolation and asymmetric registration, bias can be pronounced. Either improved interpolation or symmetric registration alone can greatly reduce this bias, and our proposed method combining both aspects shows no significant bias in any of the four experiments.
Anatomical segmentation of structures of interest is critical to quantitative analysis in medical imaging. Several automated multi-atlas based segmentation propagation methods that utilise manual delineations from multiple templates appear promising. However, high levels of accuracy and reliability are needed for use in diagnosis or in clinical trials. We propose a new local ranking strategy for template selection based on the locally normalised cross correlation (LNCC) and an extension to the classical STAPLE algorithm by Warfield et al. (2004), which we refer to as STEPS for Similarity and Truth Estimation for Propagated Segmentations. It addresses the well-known problems of local vs. global image matching and the bias introduced in the performance estimation due to structure size. We assessed the method on hippocampal segmentation using a leave-one-out cross validation with optimised model parameters; STEPS achieved a mean Dice score of 0.925 when compared with manual segmentation. This was significantly better in terms of segmentation accuracy when compared to other state-of-the-art fusion techniques. Furthermore, due to the finer anatomical scale, STEPS also obtains more accurate segmentations even when using only a third of the templates, reducing the dependence on large template databases. Using a subset of Alzheimer’s Disease Neuroimaging Initiative (ADNI) scans from different MRI imaging systems and protocols, STEPS yielded similarly accurate segmentations (Dice = 0.903). A cross-sectional and lon- gitudinal hippocampal volumetric study was performed on the ADNI database. Mean ± SD hippocampal volume (mm3 ) was 5195 ± 656 for controls; 4786 ± 781 for MCI; and 4427 ± 903 for Alzheimer’s disease patients and hippocampal atrophy rates (%/year) of 1.09 ± 3.0, 2.74 ± 3.5 and 4.04 ± 3.6 respectively. Statistically significant p < 10^-3 differences were found between disease groups for both hippocampal volume and volume change rates. Finally, STEPS was also applied in a multi-label segmentation propagation scenario using a leave-one-out cross validation, in order to parcellate 83 separate structures of the brain. Comparisons of STEPS with state-of-the-art multi-label fusion algorithms showed statistically significant segmentation accuracy improvements (p < 10^-4) in several key structures.
TISSUES AND CORTICAL THICKNESS
Thickness measurements of the cerebral cortex can aid diagnosis and provide valuable information about the temporal evolution of diseases such as Alzheimer's, Huntington's, and schizophrenia. Methods that measure the thickness of the cerebral cortex from in-vivo magnetic resonance (MR) images rely on an accurate segmentation of the MR data. However, segmenting the cortex in a robust and accurate way still poses a challenge due to the presence of noise, intensity non-uniformity, partial volume effects,the limited resolution of MRI and the highly convoluted shape of the cortical folds. Beginning with a well-established probabilistic segmentation model with anatomical tissue priors, we propose three post-processing refinements: a novel modification of the prior information to reduce segmentation bias; introduction of explicit partial volume classes; and a locally varying MRF-based model for enhancement of sulci and gyri. Experiments performed on a new digital phantom, on BrainWeb data and on data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) show statistically significant improvements in Dice scores and PV estimation (pb10−3) and also increased thickness estimation accuracy when compared to three well established techniques.
The extraction of thickness measurements from shapes with spherical topology has been an active area of research in medical imaging. Measuring the thickness of structures from automatic probabilistic ume (PV) effects and the limited resolution of medical images. Also, the complexity of certain shapes, like the highly convoluted and PV ments. In this paper we explore the use of Khalimsky's cubic complex for the extraction of topologically correct thickness measurements from probabilistic or fuzzy segmentations without explicit parametrisation of the edge. A sequence of element collapse operations is used to correct the topology of the segmentation. The Laplace equation is then solved between multiple equipotential lines and the thickness measured with an ordered upwind differencing method using an anisotropic grid with the probabilistic segmentation as a speed function. Experiments performed on digital phantoms show that the proposed method obtains topologically correct thickness measurements with an increase in accuracy when compared to two well established techniques. Furthermore, quantitative analysis on brain MRI data showed that the proposed algorithm is able to retrieve expected group differences between the cortical thickness of AD patients and controls with high statistical significance.
DIFFUSION WEIGHTED IMAGING (DWI/DTI)
We introduce a novel image-processing framework for tracking longitudinal changes in white matter micro- structure using diffusion tensor imaging (DTI). Charting the trajectory of such temporal changes offers new insight into disease progression but to do so accurately faces a number of challenges. Recent developments have highlighted the importance of processing each subject's data at multiple time points in an unbiased way. In this paper, we aim to highlight a different challenge critical to the processing of longitudinal DTI data, namely the approach to image alignment. Standard approaches in the literature align DTI data by registering the corresponding scalar-valued fractional anisotropy (FA) maps. We propose instead a DTI registration algorithm that leverages full tensor information to drive improved alignment. This proposed pipeline is evaluated against the standard FA-based approach using a DTI dataset from an ongoing study of Alzheimer's disease (AD). The dataset consists of subjects scanned at two time points and at each time point the DTI acquisition consists of two back-to-back repeats in the same scanning session. The repeated scans allow us to evaluate the specificity of each pipeline, using a test–retest design, and assess precision, using bootstrap-based method. The results show that the tensor-based pipeline achieves both higher specificity and precision than the standard FA-based approach. Tensor-based registration for longitudinal processing of DTI data in clinical studies may be of particular value in studies assessing disease progression.
POSITRON EMISSION TOMOGRAPHY (PET) IMAGING
There is considerable interest in designing therapeutic studies of individuals at risk of Alzheimer disease (AD) to prevent the onset of symptoms. Cortical b-amyloid plaques, the first stage of AD pathology, can be detected in vivo using positron emission tomography (PET), and several studies have shown that ,1/3 of healthy elderly have significant b-amyloid deposition. Here we assessed whether asymptomatic amyloid-PET-positive controls have increased rates of brain atrophy, which could be harnessed as an outcome measure for AD prevention trials. We assessed 66 control subjects (age=73.567.3yrs; MMSE=2961.3) from the Australian Imaging Biomarkers & Lifestyle study who had a baseline Pittsburgh Compound B (PiB) PET scan and two 3T MRI scans ,18-months apart. We calculated PET standard uptake value ratios (SUVR), and classified individuals as amyloid-positive/negative. Baseline and 18-month MRI scans were registered, and brain, hippocampal, and ventricular volumes and annualized volume changes calculated. Increasing baseline PiB-PET measures of b-amyloid load correlated with hippocampal atrophy rate independent of age (p = 0.014). Twenty-two (1/3) were PiB-positive (SUVR.1.40), the remaining 44 PiB-negative (SUVR#1.31). Compared to PiB-negatives, PiB-positive individuals were older (76.867.5 vs. 71.767.5, p,0.05) and more were APOE4 positive (63.6% vs. 19.2%, p,0.01) but there were no differences in baseline brain, ventricle or hippocampal volumes, either with or without correction for total intracranial volume, once age and gender were accounted for. The PiB-positive group had greater total hippocampal loss (0.0660.08 vs. 0.0260.05 ml/yr, p = 0.02), independent of age and gender, with non-significantly higher rates of whole brain (7.169.4 vs. 4.765.5 ml/yr) and ventricular (2.063.0 vs. 1.161.0 ml/yr) change. Based on the observed effect size, recruiting 384 (95%CI 195–1080) amyloid-positive subjects/arm will provide 80% power to detect 25% absolute slowing of hippocampal atrophy rate in an 18-month treatment trial. We conclude that hippocampal atrophy may be a feasible outcome measure for secondary prevention studies in asymptomatic amyloidosis.
FUNCTIONAL MRI (fMRI)
The human cortex is organised in a dense network of intricate connectvity. Neurodegenerative diseases disrupt functional connectivity by targeting key areas in this network. In Alzheimer's disease, the disruption of the Default-Mode Network constitutes a prevalent finding in many patients, and therefore constitutes a very promising functional biomarker for Alzheimer's disease. Multiple imaging modalities including neasurement of glucose metabolism (see PET Imaging), intrinisc/task-evoked brain activity (fMRI) and structural atrophy (see global and local atrophy) indicate dysfunction of the DMN in Alzheimer patients.
Functional MRI modalities such as BOLD-weighted fMRI or ASL are influenced by many confounds that interefere with the measured signal. Especially physiological influences including respiration, cardiac cycle, vasomotor oscillations, blood pressure an cerebral autoregulation, as well as head motion can bias or superimpose the actual signal of interest. The development of techniques that address the removal of such interfering variance is thus of great importance for the analysis.
There has been a great deal of recent work making use of multivariate classification techniques such as support vector machines to classify brain images obtained from MRI or PET as healthy or suffering from neurodegenerative disease. In the case of Alzheimer’s disease, the results are as accurate as standard clinical tests and could potentially be used in a diagnostic setting. However these techniques give categorical class decisions. Here we show for the first time that Gaussian processes can be applied to structural neuroimaging data to perform classification of Alzheimer’s disease subjects in a fully Bayesian framework. This offers advantages such as automatic setting of parameters via type II maximum likelihood and probabilistic predictions that may be useful in a clinical context, while maintaining the same accuracy as a state-of-the-art discriminative classifier applied to the same data.