Andrea Rueda

Single-image super-resolution of brain MR images using overcomplete dictionaries

Resolution in Magnetic Resonance (MR) is limited by diverse physical, technological and economical considerations. In conventional medical practice, resolution enhancement is usually performed with bicubic or B-spline interpolations, strongly affecting the accuracy of subsequent processing steps such as segmentation or registration. This paper presents a sparse-based super-resolution method, adapted for easily including prior knowledge, which couples up high and low frequency information so that a high-resolution version of a low-resolution brain MR image is generated. The proposed approach includes a whole-image multi-scale edge analysis and a dimensionality reduction scheme, which results in a remarkable improvement of the computational speed and accuracy, taking nearly 26 min to generate a complete 3D high-resolution reconstruction. The method was validated by comparing interpolated and reconstructed versions of 29 MR brain volumes with the original images, acquired in a 3T scanner, obtaining a reduction of 70% in the root mean squared error, an increment of 10.3 dB in the peak signal-to-noise ratio, and an agreement of 85% in the binary gray matter segmentations. The proposed method is shown to outperform a recent state-of-the-art algorithm, suggesting a substantial impact in voxel-based morphometry studies.

Topology-corrected segmentation and local intensity estimates for improved partial volume classification of brain cortex in MRI.

In magnetic resonance imaging (MRI), accuracy and precision with which brain structures may be quantified are frequently affected by the partial volume (PV) effect. PV is due to the limited spatial resolution of MRI compared to the size of anatomical structures. Accurate classification of mixed voxels and correct estimation of the proportion of each pure tissue (fractional content) may help to increase the precision of cortical thickness estimation in regions where this measure is particularly difficult, such as deep sulci. The contribution of this work is twofold: on the one hand, we propose a new method to label voxels and compute tissue fractional content, integrating a mechanism for detecting sulci with topology preserving operators. On the other hand, we improve the computation of the fractional content of mixed voxels using local estimation of pure tissue intensity means. Accuracy and precision were assessed using simulated and real MR data and comparison with other existing approaches demonstrated the benefits of our method. Significant improvements in gray matter (GM) classification and cortical thickness estimation were brought by the topology correction. The fractional content root mean squared error diminished by 6.3% (p<0.01) on simulated data. The reproducibility error decreased by 8.8% (p<0.001) and the Jaccard similarity measure increased by 3.5% on real data. Furthermore, compared with manually guided expert segmentations, the similarity measure was improved by 12.0% (p<0.001). Thickness estimation with the proposed method showed a higher reproducibility compared with the measure performed after partial volume classification using other methods.

Extracting Salient Brain Patterns for Imaging-Based Classification of Neurodegenerative Diseases

Neurodegenerative diseases comprise a wide variety of mental symptoms whose evolution is not directly related to the visual analysis made by radiologists, who can hardly quantify systematic differences. Moreover, automatic brain morphometric analyses, that do perform this quantification, contribute very little to the comprehension of the disease, i.e., many of these methods classify but they do not produce useful anatomo-functional correlations. This paper presents a new fully automatic image analysis method that reveals discriminative brain patterns associated to the presence of neurodegenerative diseases, mining systematic differences and therefore grading objectively any neurological disorder. This is accomplished by a fusion strategy that mixes together bottom-up and top-down information flows. Bottom-up information comes from a multiscale analysis of different image features, while the top-down stage includes learning and fusion strategies formulated as a max-margin multiple-kernel optimization problem. The capacity of finding discriminative anatomic patterns was evaluated using the Alzheimers disease (AD) as the use case. The classification performance was assessed under different configurations of the proposed approach in two public brain magnetic resonance datasets (OASIS-MIRIAD) with patients diagnosed with AD, showing an improvement varying from 6.2% to 13% in the equal error rate measure, with respect to what has been reported by the feature-based morphometry strategy. In terms of the anatomical analysis, discriminant regions found by the proposed approach highly correlates to what has been reported in clinical studies of AD.

Bag of Features for Automatic Classification of Alzheimer’s Disease in Magnetic Resonance Images

The goal of this paper is to evaluate the suitability of a bag-of-feature representation for automatic classification of Alzheimer’s disease brain magnetic resonance (MR) images. The evaluated method uses a bag-of-features (BOF) to represent the MR images, which are then fed to a support vector machine, which has been trained to distinguish between normal control and Alzheimer’s disease. The method was applied to a set of images from the OASIS data set. An exhaustive exploration of different BOF parameters was performed, i.e. feature extraction, dictionary construction and classification model. The experimental results show that the evaluated method reaches competitive performance in terms of accuracy, sensibility and specificity. In particular, the method based on a BOF representation outperforms the best published result in this data set improving the equal error classification rate in about 10% (0.80 to 0.95 for Group 1 and 0.71 to 0.81 for Group 2).

Cortical surface mapping using topology correction, partial flattening and 3D shape context-based non-rigid registration for use in quantifying atrophy in Alzheimer's disease

Magnetic resonance (MR) provides a non-invasive way to investigate changes in the brain resulting from aging or neurodegenerative disorders such as Alzheimers disease (AD). Performing accurate analysis for population studies is challenging because of the interindividual anatomical variability. A large set of tools is found to perform studies of brain anatomy and population analysis (FreeSurfer, SPM, FSL). In this paper we present a newly developed surface-based processing pipeline (MILXCTE) that allows accurate vertex-wise statistical comparisons of brain modifications, such as cortical thickness (CTE). The brain is first segmented into the three main tissues: white matter, gray matter and cerebrospinal fluid, after CTE is computed, a topology corrected mesh is generated. Partial inflation and non-rigid registration of cortical surfaces to a common space using shape context are then performed. Each of the steps was firstly validated using MR images from the OASIS database. We then applied the pipeline to a sample of individuals randomly selected from the AIBL study on AD and compared with FreeSurfer. For a population of 50 individuals we found correlation of cortical thickness in all the regions of the brain (average r=0.62 left and r=0.64 right hemispheres). We finally computed changes in atrophy in 32 AD patients and 81 healthy elderly individuals. Significant differences were found in regions known to be affected in AD. We demonstrated the validity of the method for use in clinical studies which provides an alternative to well established techniques to compare different imaging biomarkers for the study of neurodegenerative diseases.

3D shape context surface registration for cortical mapping

Deformable registration of cortical surfaces facilitates longitudinal and intergroup comparisons of cortical structure and function in the study of many neurodegenerative diseases. Non-rigid cortical matching is a challenging task due to the large variability between individuals and the complexity of the cortex. We present a new framework for computing cortical correspondences on brain surfaces based on 3D Shape Context and mean curvatures of partially flattened surfaces (PFS). Our approach is scale invariant and provides an accurate and anatomically meaningful alignment across the population. Registering PFS, instead of original cortical surfaces, simplifies the determination of shape correspondences, overcoming the problem of intersubject variability, while still guaranteeing the alignment of the main brain lobes and folding patterns. We validated the approach using 30 segmented brains from the OASIS database registered to a common space and compared the results with Freesurfer. In average, mean absolute distance of 0.36 and Hausdorff distance of 5.06 between moving and target surfaces are obtained. Further localization of labelled areas on each hemisphere demonstrated the accuracy of the technique.

Partial volume estimation of brain cortex from MRI using topology-corrected segmentation

In magnetic resonance imaging (MRI), accuracy of brain structures quantification may be affected by the partial volume (PV) effect. PV is due to the limited spatial resolution of MRI compared to the size of anatomical structures. When considering the cortex, measurements can be even more difficult as it spans only a few voxels. In tight sulci areas, where the two banks of the cortex are in contact, voxels may be misclassified. The aim of this work is to propose a new PV classification-estimation method which integrates a mechanism for correcting sulci delineation using topology preserving operators after a maximum a posteriori classification. Additionally, we improved the estimation of mixed voxels fractional content by adaptively estimating pure tissue intensity means. Accuracy and precision were assessed using simulated and real MR data and comparison with other existing approaches demonstrated the benefits of our method. Significant improvements in GM classification were brought by the topology correction. The root mean squared error diminished by 6.3% (p ≪ 0.01) on simulated data. The reproducibility error decreased by 9.6% (p ≪ 0.001) and the similarity measure (Jaccard) increased by 3.4% on real data. Furthermore, compared with manually-guided expert segmentations the similarity measure was improved by 12.0% (p ≪ 0.001).

Discovering Regional Pathological Patterns in Brain MRI

Complex pathological brain patterns generally are found in neurodegenerative diseases which can be correlated with different clinical onsets of a particular pathology. Currently, an objective method that aids to determine such signs, in terms of global and local changes, is not available in clinical practice and the whole interpretation is dependent on the radiologists skills. In this paper, we propose a fully automatic method that analyzes the brain structure under a multidimensional frame and highlights relevant brain patterns. An association of such patterns with the disease is herein evaluated in three classification tasks, involving probable Alzheimers disease (AD) patients, Mild Cognitive Impairment (MCI) patients and normal subjects (NC). A set of 75 brain MR images from NC subjects (25), MCI (25) and probable AD (25) patients, split into training (15 subjects) and testing (60 subjects) sets, was used to evaluate the performance of the proposed approach. Preliminary results show that the proposed method reaches a maximum classification accuracy of 80% when discriminating AD patients from NC, of 75% for classification of MCI patients from NC.

Classification of Alzheimer's disease using regional saliency maps from brain MR volumes

Accurate diagnosis of Alzheimers disease (AD) from structural Magnetic Resonance (MR) images is difficult due to the complex alteration of patterns in brain anatomy that could indicate the presence or absence of the pathology. Currently, an effective approach that allows to interpret the disease in terms of global and local changes is not available in the clinical practice. In this paper, we propose an approach for classification of brain MR images, based on finding pathology-related patterns through the identification of regional structural changes. The approach combines a probabilistic Latent Semantic Analysis (pLSA) technique, which allows to identify image regions through latent topics inferred from the brain MR slices, with a bottom-up Graph-Based Visual Saliency (GBVS) model, which calculates maps of relevant information per region. Regional saliency maps are finally combined into a single map on each slice, obtaining a master saliency map of each brain volume. The proposed approach includes a one-to-one comparison of the saliency maps which feeds a Support Vector Machine (SVM) classifier, to group test subjects into normal or probable AD subjects. A set of 156 brain MR images from healthy (76) and pathological (80) subjects, splitted into a training set (10 non-demented and 10 demented subjects) and one testing set (136 subjects), was used to evaluate the performance of the proposed approach. Preliminary results show that the proposed method reaches a maximum classification accuracy of 87.21%.

A sparse Bayesian representation for super-resolution of cardiac MR images

High-quality cardiac magnetic resonance (CMR) images can be hardly obtained when intrinsic noise sources are present, namely heart and breathing movements. Yet heart images may be acquired in real time, the image quality is really limited and most sequences use ECG gating to capture images at each stage of the cardiac cycle during several heart beats. This paper presents a novel super-resolution algorithm that improves the cardiac image quality using a sparse Bayesian approach. The high-resolution version of the cardiac image is constructed by combining the information of the low-resolution series -observations from different non-orthogonal series composed of anisotropic voxels - with a prior distribution of the high-resolution local coefficients that enforces sparsity. In addition, a global prior, extracted from the observed data, regularizes the solution. Quantitative and qualitative validations were performed in synthetic and real images w.r.t to a baseline, showing an average increment between 2.8 and 3.2 dB in the Peak Signal-to-Noise Ratio (PSNR), between 1.8% and 2.6% in the Structural Similarity Index (SSIM) and 2.% to 4% in quality assessment (IL-NIQE). The obtained results demonstrated that the proposed method is able to accurately reconstruct a cardiac image, recovering the original shape with less artifacts and low noise.