Kelvin K. Leung

Automated cross-sectional and longitudinal hippocampal volume measurement in mild cognitive impairment and Alzheimer's disease

Volume and change in volume of the hippocampus are both important markers of Alzheimers disease (AD). Delineation of the structure on MRI is time-consuming and therefore reliable automated methods are required. We describe an improvement (multiple-atlas propagation and segmentation (MAPS)) to our template library-based segmentation technique. The improved technique uses non-linear registration of the best-matched templates from our manually segmented library to generate multiple segmentations and combines them using the simultaneous truth and performance level estimation (STAPLE) algorithm. Change in volume over 12months (MAPS-HBSI) was measured by applying the boundary shift integral using MAPS regions. Methods were developed and validated against manual measures using subsets from Alzheimers Disease Neuroimaging Initiative (ADNI). The best method was applied to 682 ADNI subjects, at baseline and 12-month follow-up, enabling assessment of volumes and atrophy rates in control, mild cognitive impairment (MCI) and AD groups, and within MCI subgroups classified by subsequent clinical outcome. We compared our measures with those generated by Surgical Navigation Technologies (SNT) available from ADNI. The accuracy of our volumes was one of the highest reported (mean(SD) Jaccard Index 0.80(0.04) (N=30)). Both MAPS baseline volume and MAPS-HBSI atrophy rate distinguished between control, MCI and AD groups. Comparing MCI subgroups (reverters, stable and converters): volumes were lower and rates higher in converters compared with stable and reverter groups (p< or =0.03). MAPS-HBSI required the lowest sample sizes (78 subjects) for a hypothetical trial. In conclusion, the MAPS and MAPS-HBSI methods give accurate and reliable volumes and atrophy rates across the clinical spectrum from healthy aging to AD.

BEaST: brain extraction based on nonlocal segmentation technique.

Brain extraction is an important step in the analysis of brain images. The variability in brain morphology and the difference in intensity characteristics due to imaging sequences make the development of a general purpose brain extraction algorithm challenging. To address this issue, we propose a new robust method (BEaST) dedicated to produce consistent and accurate brain extraction. This method is based on nonlocal segmentation embedded in a multi-resolution framework. A library of 80 priors is semi-automatically constructed from the NIH-sponsored MRI study of normal brain development, the International Consortium for Brain Mapping, and the Alzheimers Disease Neuroimaging Initiative databases. In testing, a mean Dice similarity coefficient of 0.9834±0.0053 was obtained when performing leave-one-out cross validation selecting only 20 priors from the library. Validation using the online Segmentation Validation Engine resulted in a top ranking position with a mean Dice coefficient of 0.9781±0.0047. Robustness of BEaST is demonstrated on all baseline ADNI data, resulting in a very low failure rate. The segmentation accuracy of the method is better than two widely used publicly available methods and recent state-of-the-art hybrid approaches. BEaST provides results comparable to a recent label fusion approach, while being 40 times faster and requiring a much smaller library of priors.

STEPS: Similarity and Truth Estimation for Propagated Segmentations and its application to hippocampal segmentation and brain parcelation

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 Alzheimers Disease Neuroimaging Initiative (ADNI) scans from different MRI imaging systems and protocols, STEPS yielded similarly accurate segmentations (Dice=0.903). A cross-sectional and longitudinal hippocampal volumetric study was performed on the ADNI database. Mean±SD hippocampal volume (mm(3)) was 5195 ± 656 for controls; 4786 ± 781 for MCI; and 4427 ± 903 for Alzheimers 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.

Accurate automatic estimation of total intracranial volume: a nuisance variable with less nuisance.

Total intracranial volume (TIV/ICV) is an important covariate for volumetric analyses of the brain and brain regions, especially in the study of neurodegenerative diseases, where it can provide a proxy of maximum pre-morbid brain volume. The gold-standard method is manual delineation of brain scans, but this requires careful work by trained operators. We evaluated Statistical Parametric Mapping 12 (SPM12) automated segmentation for TIV measurement in place of manual segmentation and also compared it with SPM8 and FreeSurfer 5.3.0. For T1-weighted MRI acquired from 288 participants in a multi-centre clinical trial in Alzheimers disease we find a high correlation between SPM12 TIV and manual TIV (R2 = 0.940, 95% Confidence Interval (0.924, 0.953)), with a small mean difference (SPM12 40.4 ± 35.4 ml lower than manual, amounting to 2.8% of the overall mean TIV in the study). The correlation with manual measurements (the key aspect when using TIV as a covariate) for SPM12 was significantly higher (p < 0.001) than for either SPM8 (R2 = 0.577 CI (0.500, 0.644)) or FreeSurfer (R2 = 0.801 CI (0.744, 0.843)). These results suggest that SPM12 TIV estimates are an acceptable substitute for labour-intensive manual estimates even in the challenging context of multiple centres and the presence of neurodegenerative pathology. We also briefly discuss some aspects of the statistical modelling approaches to adjust for TIV.

Robust atrophy rate measurement in Alzheimer's disease using multi-site serial MRI: Tissue-specific intensity normalization and parameter selection

We describe an improved method of measuring brain atrophy rates from serial MRI for multi-site imaging studies of Alzheimers disease (AD). The method (referred to as KN-BSI) improves an existing brain atrophy measurement technique-the boundary shift integral (classic-BSI), by performing tissue-specific intensity normalization and parameter selection. We applied KN-BSI to measure brain atrophy rates of 200 normal and 141 AD subjects using baseline and 1-year MRI scans downloaded from the Alzheimers Disease Neuroimaging Initiative database. Baseline and repeat images were reviewed as pairs by expert raters and given quality scores. Including all image pairs, regardless of quality score, mean KN-BSI atrophy rates were 0.09% higher (95% CI 0.03% to 0.16%, p=0.007) than classic-BSI rates in controls and 0.07% higher (-0.01% to 0.16%, p=0.07) higher in ADs. The SD of the KN-BSI rates was 22% lower (15% to 29%, p<0.001) in controls and 13% lower (6% to 20%, p=0.001) in ADs, compared to classic-BSI. Using these results, the estimated sample size (needed per treatment arm) for a hypothetical trial of a treatment for AD (80% power, 5% significance to detect a 25% reduction in atrophy rate) would be reduced from 120 to 81 (a 32% reduction, 95% CI=18% to 45%, p<0.001) when using KN-BSI instead of classic-BSI. We concluded that KN-BSI offers more robust brain atrophy measurement than classic-BSI and substantially reduces sample sizes needed in clinical trials.

Brain MAPS: An automated, accurate and robust brain extraction technique using a template library.

Whole brain extraction is an important pre-processing step in neuroimage analysis. Manual or semi-automated brain delineations are labour-intensive and thus not desirable in large studies, meaning that automated techniques are preferable. The accuracy and robustness of automated methods are crucial because human expertise may be required to correct any suboptimal results, which can be very time consuming. We compared the accuracy of four automated brain extraction methods: Brain Extraction Tool (BET), Brain Surface Extractor (BSE), Hybrid Watershed Algorithm (HWA) and a Multi-Atlas Propagation and Segmentation (MAPS) technique we have previously developed for hippocampal segmentation. The four methods were applied to extract whole brains from 682 1.5T and 157 3T T(1)-weighted MR baseline images from the Alzheimers Disease Neuroimaging Initiative database. Semi-automated brain segmentations with manual editing and checking were used as the gold-standard to compare with the results. The median Jaccard index of MAPS was higher than HWA, BET and BSE in 1.5T and 3T scans (p<0.05, all tests), and the 1st to 99th centile range of the Jaccard index of MAPS was smaller than HWA, BET and BSE in 1.5T and 3T scans ( p<0.05, all tests). HWA and MAPS were found to be best at including all brain tissues (median false negative rate ≤0.010% for 1.5T scans and ≤0.019% for 3T scans, both methods). The median Jaccard index of MAPS were similar in both 1.5T and 3T scans, whereas those of BET, BSE and HWA were higher in 1.5T scans than 3T scans (p<0.05, all tests). We found that the diagnostic group had a small effect on the median Jaccard index of all four methods. In conclusion, MAPS had relatively high accuracy and low variability compared to HWA, BET and BSE in MR scans with and without atrophy.

A comparison of voxel and surface based cortical thickness estimation methods

Cortical thickness estimation performed in-vivo via magnetic resonance imaging is an important technique for the diagnosis and understanding of the progression of neurodegenerative diseases. Currently, two different computational paradigms exist, with methods generally classified as either surface or voxel-based. This paper provides a much needed comparison of the surface-based method FreeSurfer and two voxel-based methods using clinical data. We test the effects of computing regional statistics using two different atlases and demonstrate that this makes a significant difference to the cortical thickness results. We assess reproducibility, and show that FreeSurfer has a regional standard deviation of thickness difference on same day scans that is significantly lower than either a Laplacian or Registration based method and discuss the trade off between reproducibility and segmentation accuracy caused by bending energy constraints. We demonstrate that voxel-based methods can detect similar patterns of group-wise differences as well as FreeSurfer in typical applications such as producing group-wise maps of statistically significant thickness change, but that regional statistics can vary between methods. We use a Support Vector Machine to classify patients against controls and did not find statistically significantly different results with voxel based methods compared to FreeSurfer. Finally we assessed longitudinal performance and concluded that currently FreeSurfer provides the most plausible measure of change over time, with further work required for voxel based methods.

Cerebral atrophy in mild cognitive impairment and Alzheimer disease Rates and acceleration

Objective: To quantify the regional and global cerebral atrophy rates and assess acceleration rates in healthy controls, subjects with mild cognitive impairment (MCI), and subjects with mild Alzheimer disease (AD). Methods: Using 0-, 6-, 12-, 18-, 24-, and 36-month MRI scans of controls and subjects with MCI and AD from the Alzheimers Disease Neuroimaging Initiative (ADNI) database, we calculated volume change of whole brain, hippocampus, and ventricles between all pairs of scans using the boundary shift integral. Results: We found no evidence of acceleration in whole-brain atrophy rates in any group. There was evidence that hippocampal atrophy rates in MCI subjects accelerate by 0.22%/year2 on average (p = 0.037). There was evidence of acceleration in rates of ventricular enlargement in subjects with MCI (p = 0.001) and AD (p < 0.001), with rates estimated to increase by 0.27 mL/year2 (95% confidence interval 0.12, 0.43) and 0.88 mL/year2 (95% confidence interval 0.47, 1.29), respectively. A post hoc analysis suggested that the acceleration of hippocampal loss in MCI subjects was mainly driven by the MCI subjects that were observed to progress to clinical AD within 3 years of baseline, with this group showing hippocampal atrophy rate acceleration of 0.50%/year2 (p = 0.003). Conclusions: The small acceleration rates suggest a long period of transition to the pathologic losses seen in clinical AD. The acceleration in hippocampal atrophy rates in MCI subjects in the ADNI seems to be driven by those MCI subjects who concurrently progressed to a clinical diagnosis of AD.

Comparison of phantom and registration scaling corrections using the ADNI cohort

Rates of brain atrophy derived from serial magnetic resonance (MR) studies may be used to assess therapies for Alzheimers disease (AD). These measures may be confounded by changes in scanner voxel sizes. For this reason, the Alzheimers Disease Neuroimaging Initiative (ADNI) included the imaging of a geometric phantom with every scan. This study compares voxel scaling correction using a phantom with correction using a 9 degrees of freedom (9DOF) registration algorithm. We took 129 pairs of baseline and 1-year repeat scans, and calculated the volume scaling correction, previously measured using the phantom. We used the registration algorithm to quantify any residual scaling errors, and found the algorithm to be unbiased, with no significant (p=0.97) difference between control (n=79) and AD subjects (n=50), but with a mean (SD) absolute volume change of 0.20 (0.20) % due to linear scalings. 9DOF registration was shown to be comparable to geometric phantom correction in terms of the effect on atrophy measurement and unbiased with respect to disease status. These results suggest that the additional expense and logistic effort of scanning a phantom with every patient scan can be avoided by registration-based scaling correction. Furthermore, based upon the atrophy rates in the AD subjects in this study, sample size requirements would be approximately 10-12% lower with (either) correction for voxel scaling than if no correction was used.

Longitudinal neuroimaging and neuropsychological profiles of frontotemporal dementia with C9ORF72 expansions

IntroductionFrontotemporal dementia (FTD) is a common cause of early-onset dementia with a significant genetic component, as underlined by the recent identification of repeat expansions in the gene C9ORF72 as a major cause of FTD and motor neuron disease. Understanding the neurobiology and clinical phenomenology of this novel mutation is currently a major research focus. However, few data are available concerning the longitudinal evolution of this genetic disease. Here we present longitudinal neuropsychological and neuroimaging data on a cohort of patients with pathological repeat expansions in C9ORF72.MethodsFollowing a review of the University College London FTD DNA database, 20 cases were retrospectively identified with a C9ORF72 expansion. Twelve cases had longitudinal neuropsychology data available and six of these cases also had longitudinal volumetric brain magnetic resonance imaging. Cortical and subcortical volumes were extracted using FreeSurfer. Rates of whole brain, hemispheric, cerebellar and ventricular change were calculated for each subject. Nonlinear fluid registration of follow-up to baseline scan was performed to visualise longitudinal intra-subject patterns of brain atrophy and ventricular expansion.ResultsPatients had low average verbal and performance IQ at baseline that became impaired (< 5th percentile) at follow-up. In particular, visual memory, naming and dominant parietal skills all showed deterioration. Mean rates of whole brain atrophy (1.4%/year) and ventricular expansion (3.2 ml/year) were substantially greater in patients with the C9ORF72 mutation than in healthy controls; atrophy was symmetrical between the cerebral hemispheres within the C9ORF72 mutation group. The thalamus and cerebellum showed significant atrophy whereas no cortical areas were preferentially affected. Longitudinal fluid imaging in individual patients demonstrated heterogeneous patterns of progressive volume loss; however, ventricular expansion and cerebellar volume loss were consistent findings.ConclusionDisease evolution in C9ORF72-associated FTD is linked neuropsychologically with increasing involvement of parietal and amnestic functions, and neuroanatomically with rather diffuse and variable cortical and central atrophy but more consistent involvement of the cerebellum and thalamus. These longitudinal profiles are consistent with disease spread within a distributed subcortical network and demonstrate the feasibility of longitudinal biomarkers for tracking the evolution of the C9ORF72 mutation phenotype.