Marc Modat

Fast free-form deformation using graphics processing units

A large number of algorithms have been developed to perform non-rigid registration and it is a tool commonly used in medical image analysis. The free-form deformation algorithm is a well-established technique, but is extremely time consuming. In this paper we present a parallel-friendly formulation of the algorithm suitable for graphics processing unit execution. Using our approach we perform registration of T1-weighted MR images in less than 1 min and show the same level of accuracy as a classical serial implementation when performing segmentation propagation. This technology could be of significant utility in time-critical applications such as image-guided interventions, or in the processing of large data sets.

Evaluation of Registration Methods on Thoracic CT: The EMPIRE10 Challenge

EMPIRE10 (Evaluation of Methods for Pulmonary Image REgistration 2010) is a public platform for fair and meaningful comparison of registration algorithms which are applied to a database of intra patient thoracic CT image pairs. Evaluation of nonrigid registration techniques is a nontrivial task. This is compounded by the fact that researchers typically test only on their own data, which varies widely. For this reason, reliable assessment and comparison of different registration algorithms has been virtually impossible in the past. In this work we present the results of the launch phase of EMPIRE10, which comprised the comprehensive evaluation and comparison of 20 individual algorithms from leading academic and industrial research groups. All algorithms are applied to the same set of 30 thoracic CT pairs. Algorithm settings and parameters are chosen by researchers expert in the con figuration of their own method and the evaluation is independent, using the same criteria for all participants. All results are published on the EMPIRE10 website (http://empire10.isi.uu.nl). The challenge remains ongoing and open to new participants. Full results from 24 algorithms have been published at the time of writing. This paper details the organization of the challenge, the data and evaluation methods and the outcome of the initial launch with 20 algorithms. The gain in knowledge and future work are discussed.

Patterns of cortical thinning in the language variants of frontotemporal lobar degeneration

Background: Frontotemporal lobar degeneration (FTLD) is a clinically, genetically, and pathologically heterogeneous neurodegenerative disorder. Two subtypes commonly present with a language disorder: semantic dementia (SemD) and progressive nonfluent aphasia (PNFA). Methods: Patients meeting consensus criteria for PNFA and SemD who had volumetric MRI of sufficient quality to allow cortical thickness analysis were recruited from a tertiary referral clinic: 44 (11 pathologically confirmed) patients with SemD and 32 (4 pathologically confirmed) patients with PNFA and 29 age-matched and gender-matched healthy controls were recruited. Cortical thickness analysis was performed using the Freesurfer software tools. Results: Patients with SemD had significant cortical thinning in the left temporal lobe, particularly temporal pole, entorhinal cortex, and parahippocampal, fusiform, and inferior temporal gyri. A similar but less extensive pattern of loss was seen in the right temporal lobe and (with increasing severity) also in left orbitofrontal, inferior frontal, insular, and cingulate cortices. Patients with PNFA had involvement particularly of the left superior temporal lobe, inferior frontal lobe, and insula, and (with increasing severity) other areas in the left frontal, lateral temporal, and anterior parietal lobes. Similar patterns were seen in the pathologically confirmed cases. Patterns of cortical thinning differed between groups: SemD had significantly more cortical thinning in the temporal lobes bilaterally while PNFA had significantly more thinning in the frontal and parietal lobes. Conclusions: The language variants of frontotemporal lobar degeneration have distinctive and significantly different patterns of cortical thinning. Increasing disease severity is associated with spread of cortical thinning and the pattern of spread is consistent with progression of clinical deficits.

Distinct profiles of brain atrophy in frontotemporal lobar degeneration caused by progranulin and tau mutations.

Neural network breakdown is a key issue in neurodegenerative disease, but the mechanisms are poorly understood. Here we investigated patterns of brain atrophy produced by defined molecular lesions in the two common forms of genetically mediated frontotemporal lobar degeneration (FTLD). Nine patients with progranulin (GRN) mutations and eleven patients with microtubule-associated protein tau (MAPT) mutations had T1 MR brain imaging. Brain volumetry and grey and white matter voxel-based morphometry (VBM) were used to assess patterns of cross-sectional atrophy in the two groups. In a subset of patients with longitudinal MRI rates of whole-brain atrophy were derived using the brain-boundary-shift integral and a VBM-like analysis of voxel-wise longitudinal volume change was performed. The GRN mutation group showed asymmetrical atrophy whereas the MAPT group showed symmetrical atrophy. Brain volumes were smaller in the GRN group with a faster rate of whole-brain atrophy. VBM delineated a common anterior cingulate–prefrontal–insular pattern of atrophy in both disease groups. Additional disease-specific profiles of grey and white matter loss were identified on both cross-sectional and longitudinal imaging: GRN mutations were associated with asymmetrical inferior frontal, temporal and inferior parietal lobe grey matter atrophy and involvement of long intrahemispheric association white matter tracts, whereas MAPT mutations were associated with symmetrical anteromedial temporal lobe and orbitofrontal grey matter atrophy and fornix involvement. The findings suggest that the effects of GRN and MAPT mutations are expressed in partly overlapping but distinct anatomical networks that link specific molecular dysfunction with clinical phenotype.

The importance of appropriate partial volume correction for PET quantification in Alzheimer's disease.

PurposeAlzheimer’s disease (AD) is the most common form of dementia. Clinically, it is characterized by progressive cognitive and functional impairment with structural hallmarks of cortical atrophy and ventricular expansion. Amyloid plaque aggregation is also known to occur in AD subjects. In-vivo imaging of amyloid plaques is now possible with positron emission tomography (PET) radioligands. PET imaging suffers from a degrading phenomenon known as the partial volume effect (PVE). The quantitative accuracy of PET images is reduced by PVEs primarily due to the limited spatial resolution of the scanner. The degree of PVE is influenced by structure size, with smaller structures tending to suffer from more severe PVEs such as atrophied grey matter regions. The aims of this paper were to investigate the effect of partial volume correction (PVC) on the quantification of amyloid PET and to highlight the importance of selecting an appropriate PVC technique.MethodsAn improved PVC technique, region-based voxel-wise (RBV) correction, was compared against existing Van-Cittert (VC) and Müller-Gärtner (MG) methods using amyloid PET imaging data. Digital phantom data were produced using segmented MRI scans from a control subject and an AD subject. Typical tracer distributions were generated for each of the phantom anatomies. Also examined were 70 clinical PET scans acquired using [18F]flutemetamol. Volume of interest (VOI) analysis was performed for corrected and uncorrected images.ResultsPVC was shown to improve the quantitative accuracy of regional analysis performed on amyloid PET images. Of the corrections applied, VC deconvolution demonstrated the worst recovery of grey matter values. MG PVC was shown to induce biases in some grey matter regions due to grey matter variability. In addition, white matter variability was shown to influence the accuracy of MG PVC in cortical grey matter and also cerebellar grey matter, a typical reference region for amyloid PET normalization in sporadic AD. RBV was shown to be more accurate than MG in terms of grey matter and white matter uptake. An increase in within-group variability after PVC was observed and is believed to be a genuine, more accurate representation of the data rather than a correction-induced error. The standardized uptake value ratio (SUVR) threshold for classifying subjects as either amyloid-positive or amyloid-negative was found to be 1.64 in the uncorrected dataset, rising to 2.25 after PVC.ConclusionCare should be taken when applying PVC to amyloid PET images. Assumptions made in existing PVC strategies can induce biases that could lead to erroneous inferences about uptake in certain regions. The proposed RBV PVC technique accounts for within-compartment variability, with the potential to reduce errors of this kind.

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.

Magnetic resonance imaging evidence for presymptomatic change in thalamus and caudate in familial Alzheimer's disease.

Amyloid imaging studies of presymptomatic familial Alzheimer’s disease have revealed the striatum and thalamus to be the earliest sites of amyloid deposition. This study aimed to investigate whether there are associated volume and diffusivity changes in these subcortical structures during the presymptomatic and symptomatic stages of familial Alzheimer’s disease. As the thalamus and striatum are involved in neural networks subserving complex cognitive and behavioural functions, we also examined the diffusion characteristics in connecting white matter tracts. A cohort of 20 presenilin 1 mutation carriers underwent volumetric and diffusion tensor magnetic resonance imaging, neuropsychological and clinical assessments; 10 were symptomatic, 10 were presymptomatic and on average 5.6 years younger than their expected age at onset; 20 healthy control subjects were also studied. We conducted region of interest analyses of volume and diffusivity changes in the thalamus, caudate, putamen and hippocampus and examined diffusion behaviour in the white matter tracts of interest (fornix, cingulum and corpus callosum). Voxel-based morphometry and tract-based spatial statistics were also used to provide unbiased whole-brain analyses of group differences in volume and diffusion indices, respectively. We found that reduced volumes of the left thalamus and bilateral caudate were evident at a presymptomatic stage, together with increased fractional anisotropy of bilateral thalamus and left caudate. Although no significant hippocampal volume loss was evident presymptomatically, reduced mean diffusivity was observed in the right hippocampus and reduced mean and axial diffusivity in the right cingulum. In contrast, symptomatic mutation carriers showed increased mean, axial and in particular radial diffusivity, with reduced fractional anisotropy, in all of the white matter tracts of interest. The symptomatic group also showed atrophy and increased mean diffusivity in all of the subcortical grey matter regions of interest, with increased fractional anisotropy in bilateral putamen. We propose that axonal injury may be an early event in presymptomatic Alzheimer’s disease, causing an initial fall in axial and mean diffusivity, which then increases with loss of axonal density. The selective degeneration of long-coursing white matter tracts, with relative preservation of short interneurons, may account for the increase in fractional anisotropy that is seen in the thalamus and caudate presymptomatically. It may be owing to their dense connectivity that imaging changes are seen first in the thalamus and striatum, which then progress to involve other regions in a vulnerable neuronal network.

Attenuation Correction Synthesis for Hybrid PET-MR Scanners: Application to Brain Studies

Attenuation correction is an essential requirement for quantification of positron emission tomography (PET) data. In PET/CT acquisition systems, attenuation maps are derived from computed tomography (CT) images. However, in hybrid PET/MR scanners, magnetic resonance imaging (MRI) images do not directly provide a patient-specific attenuation map. The aim of the proposed work is to improve attenuation correction for PET/MR scanners by generating synthetic CTs and attenuation maps. The synthetic images are generated through a multi-atlas information propagation scheme, locally matching the MRI-derived patients morphology to a database of MRI/CT pairs, using a local image similarity measure. Results show significant improvements in CT synthesis and PET reconstruction accuracy when compared to a segmentation method using an ultrashort-echo-time MRI sequence and to a simplified atlas-based method.

Accurate multimodal probabilistic prediction of conversion to Alzheimer's disease in patients with mild cognitive impairment

Accurately identifying the patients that have mild cognitive impairment (MCI) who will go on to develop Alzheimers disease (AD) will become essential as new treatments will require identification of AD patients at earlier stages in the disease process. Most previous work in this area has centred around the same automated techniques used to diagnose AD patients from healthy controls, by coupling high dimensional brain image data or other relevant biomarker data to modern machine learning techniques. Such studies can now distinguish between AD patients and controls as accurately as an experienced clinician. Models trained on patients with AD and control subjects can also distinguish between MCI patients that will convert to AD within a given timeframe (MCI-c) and those that remain stable (MCI-s), although differences between these groups are smaller and thus, the corresponding accuracy is lower. The most common type of classifier used in these studies is the support vector machine, which gives categorical class decisions. In this paper, we introduce Gaussian process (GP) classification to the problem. This fully Bayesian method produces naturally probabilistic predictions, which we show correlate well with the actual chances of converting to AD within 3 years in a population of 96 MCI-s and 47 MCI-c subjects. Furthermore, we show that GPs can integrate multimodal data (in this study volumetric MRI, FDG-PET, cerebrospinal fluid, and APOE genotype with the classification process through the use of a mixed kernel). The GP approach aids combination of different data sources by learning parameters automatically from training data via type-II maximum likelihood, which we compare to a more conventional method based on cross validation and an SVM classifier. When the resulting probabilities from the GP are dichotomised to produce a binary classification, the results for predicting MCI conversion based on the combination of all three types of data show a balanced accuracy of 74%. This is a substantially higher accuracy than could be obtained using any individual modality or using a multikernel SVM, and is competitive with the highest accuracy yet achieved for predicting conversion within three years on the widely used ADNI dataset.

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.