Local MRI analysis approach in the diagnosis of early and prodromal Alzheimer's disease☆
Research highlights
► Temporal lobe atrophy in MRI images is measured with a novel, automated technique. ► Small, highly localized volumes convey all the information. ► A localized features-based index is built to classify both individuals and cohorts. ► Discrimination accuracy is comparable to whole brain analyses in literature. ► The index can tell prodromal Alzheimer's disease patients from Controls with high accuracy.
Introduction
In recent years, the early clinical signs of Alzheimer's disease (AD) have been extensively investigated, leading to the concept of amnestic Mild Cognitive Impairment (aMCI), an intermediate cognitive state between normal aging and dementia (Winblad et al., 2004). The aMCI condition is currently identified by both a reported and an objective memory impairment, either associated with a slight impairment in other cognitive areas (multi-domain aMCI) or not (single-domain aMCI).
In longitudinal studies the aMCI subjects are experienced either to convert to AD (converters) or not (non-converters). In the latter case they may remain stable in the aMCI state or they may even revert to normalcy. Therefore, aMCI is a clinically and pathologically heterogeneous state in need of effective and reliable strategies to predict the clinical evolution. When these are available, hopefully upcoming disease-modifying drugs will be administered only to the aMCI subjects with prodromal AD as diagnosed with the help of specific biomarkers (Dubois et al., 2007, Dubois et al., 2010).
Neuropsychology is the clinical cornerstone in the effort of performing a good and reliable predictive classification of the aMCI subjects. However such tests alone prove to be not completely satisfactory because of relatively low specificity between aMCI converters and non-converters and because of floor effects that sometimes make recall measures relatively insensitive to longitudinal changes.
The new proposed criteria for early diagnosis of AD suggest that one or more biomarkers should show typical findings (“supportive features”) for an aMCI subject to be diagnosed as affected by early AD (Dubois et al., 2007).
The evidence of medial temporal lobe (MTL) atrophy in magnetic resonance imaging (MRI) is probably the most easily accessed worldwide.
Studies carried out in the last decade indicate that MRI can be used to quantify regional atrophy in MCI population, distinguishing early and later preclinical stages of AD (Pruessner et al., 2000, Shen et al., 2002).
A key contribution to biomarker findings came from the Alzheimer's Disease Neuroimaging Initiative (ADNI), a large study launched in 2003 by the National Institute on Aging (NIA), the National Institute of Biomedical Imaging and Bioengineering (NIBIB), the Food and Drug Administration (FDA), private pharmaceutical companies and non-profit organizations. The primary goal of ADNI has been to test whether serial MRI, PET, other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of MCI and early AD.
The noticeable growth in the number of candidate biomarkers poses the question on which one can add more value to the routinely performed episodic memory tests.
In principle, an ideal biomarker should detect a specific pathophysiological feature of AD, not present in the healthy condition, in other primary dementias, or in confounding conditions. Besides being reliable, a biomarker should be detectable by means of procedures which must be relatively non-invasive, simple to perform, widely available and not too expensive.
Among others, various morphological brain measures performed by means of MRI, ranging from brain-wide, voxel-wise multivariate measures to the selective volume estimate of restricted regions of interest, such as the hippocampal formation, have been proposed as candidate biomarkers (Frisoni et al., 2006, Karow et al., 2010, Desikan et al., 2009). In particular, regional neuro-anatomical changes have been investigated as biomarkers for Alzheimer's disease (Holland et al., 2009, Vemuri et al., 2010), and aMCI conversion (Risacher et al., 2010).
The increased interest in such “supportive features” derived from neuroimaging is due to the improved image quality and to the development of novel computer-assisted image processing tools giving the possibility of an automated volume and shape quantification of brain structures. Moreover, the interest is also raised by the availability of large sets of imaging data collected by large multicenter studies such as the ADNI.
In the context of neuroimaging applied to AD, computer-aided analysis techniques have been proposed to discover and study biomarkers via texture changes in signal intensity (Freeborough and Fox, 1998), gray matter concentrations differences, atrophy of subcortical limbic structures (Thompson et al., 2004, Frisoni et al., 2006) and general cortical atrophy (Thompson et al., 2003, Lerch et al., 2005). The underlying assumption being, in the general case, that changes in neuropsychological or neurological functions under consideration have a morphological counterpart, detectable via structural MRI.
A growing body of literature has used machine learning methods to extract high-dimensional features of interest from MRI, on which classification functions are built to assist in clinical diagnosis of probable AD or predict future clinical status for individuals with MCI (Klöppel et al., 2008, Fan et al., 2008, Lao, 2004, Davatzikos et al., 2008).
Hippocampal atrophy analysis performed by means of three-dimensional (3D) MRI seems to obtain a rather good discriminative value (Ferreira et al., in press). However, according to the traditional approach such volumetric measurements typically rely on manual/semi-automated outlining of the hippocampal structures on serial MR images, which is time consuming and prone to inter-rater and intra-rater variability.
In order to overcome these difficulties a novel approach was proposed in a previous paper (Calvini et al., 2009). It is based on a simple, quick, and operator independent method for the automatic extraction of two regions around the hippocampus (one for each side of the brain) from an MR image. From such subimages, denoted there as hippocampal boxes (HBs) and containing both the hippocampus and the perihippocampal region, a statistical indicator was able to separate the AD, aMCI and controls cohorts with good accuracy.
We propose, as an evolution of the previous technique, a procedure which is able to find other pathology-specific volumes of interest (VOIs) where a high discrepancy exists between healthy controls and either MCI converters or AD patients. Within a given VOI, we shall see that only a small volume contributes to the cohort discrimination, that is, the relevant information for the clinical assessment is highly localized. A Classification Index (hereafter designated as CI) could then be computed and found potentially more accurate than the discrimination based on the previous HBs.
Section snippets
Materials and methods
Our procedure is meant to be fully automated and consists of five steps summarized as follows: (i) image preprocessing (noise removal, affine registration, gray-level intensity normalization), (ii) multiple VOI extraction by means of template-matching and rigid registration, (iii) feature computation and Random Forest-based (RF) feature classification, (iv) Support Vector Machine (SVM) analysis and CI computation, (v) CI validation and conversion probability estimation based on follow-up
Results
The first result comes directly from the RF classifier. For visualization purposes, a smoothed, thresholded IFM overlaid on a representative structural MRI scan is shown in Fig. 8. The plotted IFM is the superposition of all thresholded IFMs computed on the 9 VOIs. It shows that the relevant information is squeezed into some decidedly small areas within the VOIs.
Relevant locations mostly match those found in literature (Hinrichs et al., 2009, McEvoy et al., 2009, Misra et al., 2009, Morra et
Discussion and conclusions
This study proposes a computational neuro-anatomic method to quantify local patterns of brain atrophy in a large sample of cognitively normal individuals and in patients with prodromal or mild AD. The general approach presented here was validated on a trial set of a heterogeneous MCI population, and its results in terms of cohort discrimination are comparable to those found in recent works, where applied methodologies mostly concerned whole-brain pattern analysis (Karas et al., 2008, Davatzikos
Acknowledgments
This research was supported by Istituto Nazionale di Fisica Nucleare (INFN), Italy, under the project MAGIC-5 (Medical Application on a Grid Infrastructure Connection), a joint research project involving researchers from 6 different INFN sites in Italy: Genova, Torino, Pisa, Bari, Napoli and Lecce.
This research was also supported by grants to LR, PB and ME from Università degli Studi di Genova, Italy.
Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging
References (78)
- et al.
Empirical characterization of random forest variable importance measures
Comput. Stat. Data Anal.
(2008) - et al.
Automated hippocampal shape analysis predicts the onset of dementia in mild cognitive impairment
Neuroimage
(2011) - et al.
Automatic classification of patients with Alzheimer's disease from structural MRI: a comparison of ten methods using the ADNI database
Neuroimage
(2011) - et al.
Detection of prodromal Alzheimer's disease via pattern classification of magnetic resonance imaging
Neurobiol. Aging
(2008) MRI-derived entorhinal and hippocampal atrophy in incipient and very mild Alzheimer's disease
Neurobiol. Aging
(2001)- et al.
Research criteria for the diagnosis of Alzheimer's disease: revising the NINCDADRDA criteria
Lancet Neurol.
(2007) - et al.
Spatial patterns of brain atrophy in MCI patients, identified via high-dimensional pattern classification, predict subsequent cognitive decline
Neuroimage
(2008) - et al.
In vivo neuropathology of the hippocampal formation in AD: a radial mapping MR-based study
Neuroimage
(2006) - et al.
Spatially augmented LPBoosting for AD classification with evaluations on the ADNI dataset
Neuroimage
(2009) - et al.
Predictive markers for AD in a multi-modality framework: an analysis of MCI progression in the ADNI population
Neuroimage
(2011)
Morphological classification of brains via high-dimensional shape transformations and machine learning methods
Neuroimage
Analysis of regional MRI volumes and thicknesses as predictors of conversion from mild cognitive impairment to Alzheimer's disease
Neurobiol. Aging
Baseline and longitudinal patterns of brain atrophy in MCI patients, and their use in prediction of short-term conversion to AD: results from ADNI
Neuroimage
Automated mapping of hippocampal atrophy in 1-year repeat MRI data from 490 subjects with Alzheimer's disease, mild cognitive impairment, and elderly controls
Neuroimage
Early detection of Alzheimer's disease using neuroimaging
Exp. Gerontol.
Discrimination of AD and normal subjects from MRI: anatomical versus statistical regions
Neurosci. Lett.
Automated detection of brain atrophy patterns based on MRI for the prediction of Alzheimer's disease
Neuroimage
The LONI Pipeline Processing Environment
Neuroimage
Longitudinal MRI atrophy biomarkers: relationship to conversion in the ADNI cohort.
Neurobiol. Aging
Measuring size and shape of the hippocampus in MR images using a deformable shape model
Neuroimage
Advances in functional and structural MR image analysis and implementation as FSL
Neuroimage
Mapping hippocampal and ventricular change in Alzheimer disease
Neuroimage
Alzheimer's disease diagnosis in individual subjects using structural mr images: validation studies
Neuroimage
Multi-modal imaging predicts memory performance in normal aging and cognitive decline.
Neurobiol. Aging
High-dimensional pattern regression using machine learning: from medical images to continuous clinical variables
Neuroimage
Selection bias in working with the top genes in supervised classification of tissue samples
Stat. Method.
Cognitive activity and incident AD in a population-based sample of older persons
Neurology
Pattern of brain destruction in Parkinson's and Alzheimer's diseases
J. Neural Transm.
Random forest
Mach. Learn.
Estimating Generalization Error on Two-Class Datasets Using Out-of-Bag Estimates
Mach. Learn.
Automatic analysis of medial temporal lobe atrophy from structural MRIs for the early assessment of Alzheimer's disease
Med. Phys.
Simplified design of steerable pyramid filters, Circuits and Systems
Fully automatic hippocampus segmentation and classification in Alzheimer's disease and mild cognitive impairment applied on data from adni
Hippocampus
Hypothetical model of dynamic biomarkers of the Alzheimer's pathological cascade
Lancet Neurol.
Support-vector networks
Mach. Learn.
Automated MRI measures identify individuals with mild cognitive impairment and Alzheimer's disease
Brain J. Neurol.
The cortical signature of Alzheimer's disease: regionally specific cortical thinning relates to symptom severity in very mild to mild ad dementia and is detectable in asymptomatic amyloid-positive individuals
Cereb. Cortex
Longitudinal quantitative proton magnetic resonance spectroscopy of the hippocampus in Alzheimer's disease
Brain J. Neurol.
Cited by (168)
Study on the effect of extreme learning machine and its variants in differentiating Alzheimer conditions from selective regions of brain MR images
2022, Expert Systems with ApplicationsAutomated detection, selection and classification of hippocampal landmark points for the diagnosis of Alzheimer's disease
2022, Computer Methods and Programs in BiomedicineConvolutional Neural Network and Recursive Feature Elimination Based Model for the Diagnosis of Mild Cognitive Impairments
2024, Lecture Notes in Networks and Systems
- ☆
Data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (http://www.loni.ucla.edu/ADNI). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators is available at http://www.loni.ucla.edu/ADNI/Collaboration/ADNI_Authorship_list.pdf.