A novel method and software for automatically classifying Alzheimer’s disease patients by magnetic resonance imaging analysis

https://doi.org/10.1016/j.cmpb.2017.03.006Get rights and content

Highlights

  • Supervised classification of Alzheimer disease patients.

  • A novel technique for feature extraction from magnetic resonance images.

  • Combination of key points spatial position and their distribution around the patients brain.

  • Experimental evidence on real biomedical data sets.

  • The method outperforms state-of-the-art approaches in terms of classification performance.

Abstract

Background and objective

The cause of the Alzheimer’s disease is poorly understood and to date no treatment to stop or reverse its progression has been discovered. In developed countries, the Alzheimer’s disease is one of the most financially costly diseases due to the requirement of continuous treatments as well as the need of assistance or supervision with the most cognitively demanding activities as time goes by. The objective of this work is to present an automated approach for classifying the Alzheimer’s disease from magnetic resonance imaging (MRI) patient brain scans. The method is fast and reliable for a suitable and straightforward deploy in clinical applications for helping diagnosing and improving the efficacy of medical treatments by recognising the disease state of the patient.

Methods

Many features can be extracted from magnetic resonance images, but most are not suitable for the classification task. Therefore, we propose a new feature extraction technique from patients’ MRI brain scans that is based on a recent computer vision method, called Oriented FAST and Rotated BRIEF. The extracted features are processed with the definition and the combination of two new metrics, i.e., their spatial position and their distribution around the patient’s brain, and given as input to a function-based classifier (i.e., Support Vector Machines).

Results

We report the comparison with recent state-of-the-art approaches on two established medical data sets (ADNI and OASIS). In the case of binary classification (case vs control), our proposed approach outperforms most state-of-the-art techniques, while having comparable results with the others. Specifically, we obtain 100% (97%) of accuracy, 100% (97%) sensitivity and 99% (93%) specificity for the ADNI (OASIS) data set. When dealing with three or four classes (i.e., classification of all subjects) our method is the only one that reaches remarkable performance in terms of classification accuracy, sensitivity and specificity, outperforming the state-of-the-art approaches. In particular, in the ADNI data set we obtain a classification accuracy, sensitivity and specificity of 99% while in the OASIS data set a classification accuracy and sensitivity of 77% and specificity of 79% when dealing with four classes.

Conclusions

By providing a quantitative comparison on the two established data sets with many state-of-the-art techniques, we demonstrated the effectiveness of our proposed approach in classifying the Alzheimer’s disease from MRI patient brain scans.

Introduction

The Alzheimer’s Disease (AD) is one of the most common form of dementia, prevailing in people with over 65 years. In AD, normal brain tissues are degenerated and this fact can cause a reduction of memory and mental abilities. Magnetic resonance imaging does have a good ability for differentiating between soft tissues, thus providing useful information about organs like the brain. It is a tool that allows to diagnose many brain diseases including the Alzheimer’s disease. Multiple approaches successfully used MRI brain scans for abnormality/ normality Alzheimer assessment in the past [1], [2], [3], [4], [5]. Many other imaging modalities exists in order to study AD and other brain diseases, i.e., computed tomography (CT), positron emission tomography (PET) studies of cerebral metabolism with fluoro-deoxy-d-glucose (FDG), amyloid tracers such as Pittsburgh Compound-B (PiB), and [123I] FP-CIT SPECT [6], [7]. Among the plethora of methods being effectively applied in the classification of the Alzheimer’s disease, we briefly discuss those that are mostly related with the approach of this paper and with MRI.

Anandh et al. [8] propose to differentiate Mild Cognitive Impairment (MCI), Condition Normal (CN) and AD subjects from MRI scans using an approach based on Laplace-Beltrami eigenvalue shape descriptors. Laplace-Beltrami eigenvalues are infinite series of spectra that describes the intrinsic geometry of objects. Most significant Laplace–Beltrami shape descriptors are identified and their performance is analysed using linear Support Vector Machine (SVM) classifier.

Ramaniharan et al. [9] design another possible application of the Laplace–Beltrami eigenvalue shape descriptor for classifying the Alzheimer’s disease. The authors attempt to analyse the shape changes of corpus callosum using shape based Laplace-Beltrami eigenvalue features and machine learning techniques. Corpus callosum from the normal and AD T1-weighted MRI scans are segmented using the reaction diffusion level set approach and the results are validated against the ground-truth images. These values capture the shape information of corpus callosum by solving the eigenvalue problem of Laplace–Beltrami operator on the triangular meshes. The significant features are selected based on information gain ranking and subjected to classification using K-Nearest Neighbour and SVM. Furthermore, Guerrero et al. [10] propose a framework for feature extraction from low-dimensional subspaces that represent inter-subject variability. The manifold subspace is built from data-driven regions of interest. The regions are learnt via sparse regression using the mini-mental state examination score as an independent variable which correlates better with the actual disease stage than a discrete class label. The sparse regression is used to perform variable selection along with a re-sampling scheme to reduce sampling bias.

Conversely, Beheshti and Demirel [11] describe the use of t-test based feature-ranking approach as part of their feature extraction procedure, where the number of top features is determined using the Fisher criterion. The key aspect of this work is the use of a data fusion method among atrophy clusters to improve the classification performance. Additionally, Plocharski and Østergaard [12] design a technique to extract sulcal features by means of computing a sulcal medial surface for AD/CN classification. Zhou et al. [13] propose to combine MRI data with a neuropsychological test, mini-mental state examination as input to a multi-dimensional space for the classification of Alzheimer’s disease and its prodromal stages. The decisional space is constructed using those features deemed statistically significant via an elaborate feature selection [14] and ranking mechanism. Lastly, Daliri [15] proposes an approach based on feature extraction from MRI scans. The author presents an automated method for diagnosing Alzheimer’s disease from brain MRI scans, where a computer vision feature extractor, called Scale Invariant Feature Transforms (SIFT) [16], is used to extract important features for tackling the classification task. Finally, the reader may refer to following review papers that deal with the topic of feature extraction [17], [18], [19], [20], [21] and registration [7], [22], [23], [24] in imaging.

Contributions. The purpose of this work is to present an automated approach for classifying Alzheimer’s disease patients from magnetic resonance imaging brain scans. We propose to use a set of features derived from the concept of key points [16], representing relevant information of the images. The key points are extracted with a recent feature extraction technique, called Oriented FAST and Rotated BRIEF (ORB) [25]. One of the main contribution of this work is how the final set of features is obtained, defining and combining two new metrics, i.e., the spatial position of the extracted key points and their distribution around the patient’s brain. Our proposed method is fast and reliable for a straightforward deploy in clinical applications.

Availability. The complete software package and the documentation is open-source and publicly available at https://github.com/fabioprev/MRI-Classifier.git.

Section snippets

Methodology

In this work, we propose an approach for classifying magnetic resonance imaging scans based on a suitable choice of the features. Features may be specific structures in the image such as points, edges or objects, but they may also be the result of a general neighbourhood operation or feature detection applied to the image. Other examples of features are related to motion in image sequences, to shapes defined in terms of curves or boundaries between different image regions, or to properties of

Experimental evaluation

We now report the quantitative evaluation of the proposed approach described in Section 2.

Conclusions

In this work, we presented an automatic approach for classifying the Alzheimer’s disease from MRI patient brain scans. We demonstrated that the method is fast and reliable for a suitable and straightforward deploy in clinical applications for improving the efficacy of medical treatments by recognising the disease’s state of the patient. The method uses the Oriented FAST and Rotated BRIEF technique for extracting the salient features from each MRI patient brain scan. One of the main contribution

References (41)

  • F. de Vos et al.

    Combining multiple anatomical MRI measures improves Alzheimer’s disease classification

    Hum. Brain Mapp.

    (2016)
  • R. Wolz et al.

    Multi-method analysis of MRI images in early diagnostics of Alzheimer’s disease

    PLoS ONE

    (2011)
  • L. Bronge et al.

    Postmortem MRI and histopathology of white matter changes in Alzheimer brains

    Dement. Geriatr. Cogn. Disord.

    (2002)
  • D.P. Devanand et al.

    Hippocampal and entorhinal atrophy in mild cognitive impairment prediction of Alzheimer disease

    Neurology

    (2007)
  • L.O. Wahlund et al.

    A new rating scale for age-related white matter changes applicable to MRI and CT

    Stroke

    (2001)
  • K.A. Johnson et al.

    Brain imaging in Alzheimer disease

    Cold Spring Harb. Perspect. Med.

    (2012)
  • F. Oliveira et al.

    A robust computational solution for automated quantification of a specific binding ratio based on [123I]FP-CIT SPECT images

    Q. J. Nuclear Med. Mol. Imaging

    (2014)
  • K.R. Anandh et al.

    A method to differentiate mild cognitive impairment and Alzheimer in MR images using eigen value descriptors

    Med. Syst.

    (2016)
  • Q. Zhou et al.

    An optimal decisional space for the classification of Alzheimer’s disease and mild cognitive impairment

    Biomed. Eng.

    (2014)
  • M.R. Daliri

    Automated diagnosis of Alzheimer disease using the scale-invariant feature transforms in magnetic resonance images

    Med. Syst.

    (2012)
  • Cited by (40)

    • Early Alzheimer's disease diagnosis with the contrastive loss using paired structural MRIs

      2021, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      These inherent features combined with clinical medical data and genetic data were used to build a classifier to predict the changing trends of subjects at different stages [13]. F. Previtali et al. proposed a feature extraction technique from patients’ MRI brain scans [14]. Zhang et al. proposed a new multi-view clustering model called Consensus Multi-view Clustering (CMC) based on nonnegative matrix factorization for predicting the multiple stages of AD progression [15].

    • Artificial intelligence applications in medical imaging: A review of the medical physics research in Italy

      2021, Physica Medica
      Citation Excerpt :

      Structural MRI, based on high resolution T1 weighted imaging (3D-T1 MRI) has excellent contrast among soft tissues in brain, but diffusion tensor imaging (DTI) [3,114] or resting-state functional MRI (rsfMRI), are increasingly used for characterizing the brain activity. Accurate classification of Alzheimer disease (AD) and mild cognitive impairment (MCI) subjects has been achieved by combining novel topological descriptors and ML algorithms [115–121]. Volume descriptors or radiomic features [65] from MRI, alone or in combination with results from visuospatial tests were used in SVM classifiers to distinguish mild AD patients from healthy controls [76].

    • CMC: A consensus multi-view clustering model for predicting Alzheimer's disease progression

      2021, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      Liu et al. provided a multi-scale modeling variantto-function-to-network model to inquire into the causal effect of rare noncoding variants for AD [28]. Previtali et al. proposed a novel feature extraction method from brain MRI scans for classifying AD patient [41]. Vaithinathan et al. presented a new algorithm for the classification problems of AD, which belongs to a kind of texture extraction technique with T1 weighted MRI [49].

    View all citing articles on Scopus
    View full text