Computer-aided diagnosis of Alzheimer’s type dementia combining support vector machines and discriminant set of features☆
Introduction
Emission Computed Tomography (ECT) has been widely used in biomedical research and clinical practice during the last three decades. ECT differs from many other medical imaging modalities such as magnetic resonance imaging (MRI) in producing a mapping of physiological functions instead of imaging anatomical structures. In this way, tomographic radiopharmaceutical imaging provides in vivo three-dimensional maps of a pharmaceutical labeled with a gamma ray emitting radionuclide. The distribution of radionuclide concentrations are estimated from a set of projectional images acquired at many different angles around the patient [5].
Single Photon Emission Computed Tomography (SPECT) is an ECT imaging technique developed in the 1960s, but not widely used in clinical practice until the 1980s. It is a noninvasive, three-dimensional functional imaging modality that provides clinical information regarding biochemical and physiologic processes in patients. SPECT images are produced by the disintegration of the nucleus of a radioisotope that leads to the emission of a gamma photon with a random direction and uniformly distributed in the sphere surrounding the nucleus. If the photon does not suffer a collision with electrons or other particles with-in the body, its trajectory will be a straight line or “ray”. In order to discriminate the direction of incidence using a photon detector external to the patient, a physical collimation is required. Typically, a collimator is placed prior to the detector in such a manner that photons incident from all but a single direction are blocked by the plates. This guarantees that only photons incident from the desired direction will strike the photon detector. SPECT is essential for imaging the brain with either regional cerebral blood flow (rCBF) agents or brain receptors, and for imaging myocardial perfusion. For the past two decades, brain SPECT has become an important diagnostic and research tool in nuclear medicine [2], [32]. The ultimate value of this technology depends on good technique for image acquisition and proper data reconstruction [38], [50].
Alzheimer’s disease (AD) is a progressive neurodegenerative disorder first affecting memory functions and then gradually affecting all cognitive functions with behavioral impairments and eventually causing death. Its diagnosis is based on the information provided by a careful clinical examination, a thorough interview of the patient and relatives, and a neuropsychological assessment. A SPECT rCBF study is frequently used as a complimentary diagnostic tool in addition to the clinical findings [24], [43], [23]. However, in late-onset AD there are minimal perfusion in the mild stages of the disease, and age-related changes, which are frequently seen in healthy aged people, have to be discriminated from the minimal disease-specific changes. These minimal changes in the images make visual diagnosis a difficult task that requires experienced explorers. Even with this problem still unsolved, the potential of computer-aided diagnosis (CAD) has not been explored in this area.
Several approaches for designing CAD systems of the AD can be found in the literature [33], [26], [45]. The first family is based on the analysis of regions of interest (ROI) by means of some discriminant functions. The second approach is the statistical parametric mapping (SPM) [15] software tool and its numerous variants. SPM is widely used in neuroscience. It was not developed specifically to study a single image, but for comparing groups of images. SPM has been designed as a univariate approach since the classical multivariate techniques such as MANCOVA [47] require the number of observations (i.e. scans) to be greater than the number of components (i.e. voxels) of the multivariate observation. The importance of multivariate approaches is that the effects due to activations, confounding effects and error effects are assessed statistically, both in terms of effects at each voxel, and interactions among voxels [15]. On the other hand, statistical learning classification methods have not been explored in depth for AD CAD, quite possibly due to the fact that images represent large amounts of data and most imaging studies have relatively few subjects (generally < 100) [26], [46], [44].
Since their introduction in the late seventies [51], Support vector machines (SVMs) marked the beginning of a new era in the learning from examples paradigm [6], [27]. SVMs have focussed recent attention from the pattern recognition community due to a number of theoretical and computational merits derived from the Statistical Learning Theory (SLT) [52], [53] developed by Vladimir Vapnik at AT&T. Moreover, recent developments in defining and training statistical classifiers make it possible to build reliable classifiers in very small sample size problems [12] since pattern recognition systems based on SVM circumvent the curse of dimensionality, and even may find nonlinear decision boundaries for small training sets. These techniques have been successfully used in a number of applications [7], [49] including voice activity detection (VAD) [13], [14], [37], [40], [39], [56], [21], [20], content-based image retrieval [48], texture classification [29] and medical imaging diagnosis [16], [28], [57], [35].
This paper shows a complete CAD system for the early detection of the Alzheimer’ type dementia (ATD) by SPECT image classification. The proposed method combining SVM and advanced feature extraction schemes is developed with the aim of reducing the subjectivity in visual interpretation of SPECT scans by clinicians, thus improving the detection of the AD in its early stage. The paper is organized as follows. Section 2 provides a background on SVM classification. Section 3 summarizes the key ROIs where the disease becomes observable in its early stage. Section 4 shows the acquisition setup, reconstruction algorithm as well as the template-based spatial normalization techniques used for obtaining an accurate and anatomically standardized model of the functional brain activity provided by SPECT images. Section 5 defines first- and second-order statistics that are evaluated for building the classifier based on their discrimination ability. Finally, Section 6 shows the experiments that were conducted in order to evaluate the proposed SVM classifier as a diagnostic tool for the early detection of the AD.
Section snippets
Support vector machines
Support vector machines[6], [52], [53] are widely used for pattern recognition in a number of applications by its ability to learn from experimental data. The reason is that SVM often performs better than other conventional parametric classifiers [27], [19]. SVM separate a given set of binary labeled training data by means of a hyperplane that is maximally distant from the two classes (known as the maximal margin hyperplane). The objective is to build a function using training data
Diagnosis of Alzheimer’s type dementia by means of SPECT
Functional SPECT imaging providing information about the rCBF have been found to be a valuable aid for the early diagnosis of the AD [18]. Fig. 2 shows typical brain perfusion patterns of a normal subject and a patient affected by AD. Although many studies exist no final agreement has been achieved for the best regions of the brain to be quantified when diagnosing AD:
- •
Many studies have shown the temporo-parietal region to be practical for the early detection of the disease in patients that are
SPECT image acquisition and preprocessing
The ultimate value of a CAD system strongly depends on effective techniques for image acquisition, proper data reconstruction and image registration [1], [42]. After introducing all the necessary knowledge and tools for building the diagnosis system, this section shows the image acquisition setup and preprocessing steps of the SPECT scans that are needed prior to defining the classifier.
Discriminant statistics of Alzheimer’s type dementia
A major problem associated with pattern recognition systems is the so-called curse of dimensionality, that is, the number of available features for designing the classifier can be very large compared with the number of available training examples. Under these conditions, the difficulty of an estimation problem increases drastically with the dimension of the space, since one needs an exponential increasing number of patterns to sample the space properly. This well known statement induces some
Evaluation results
This section shows the experimental results carried out in order to evaluate the performance of the classification system and its utility as a CAD system for the early AD. First, a baseline system based on the voxel-as-features (VAF) [45], [46], [44] paradigm is implemented for reference. Second, the experimental results that were conducted to evaluate the proposed system are shown.
The SPECT images used in this work were initially labeled by experienced clinicians of the “Virgen de las Nieves”
Conclusions
This paper showed a fully automatic computer-aided diagnosis system for improving the early detection of the AD. The proposed approach is based on image parameter selection and support vector machine (SVM) classification. The system was developed by exploring the most discriminant set of input data including first and second-order statistics of sagittal, coronal and transversal sections of the human brain. It was found that the most discriminant image parameters of the AD are the coronal
References (57)
- et al.
Automatic identification of cardiac health using modeling techniques: a comparative study
Information Sciences
(2008) - et al.
Associated evolution of a support vector machine-based classifier for pedestrian detection
Information Sciences
(2009) - et al.
Hard C-means clustering for voice activity detection
Speech Communication Journal
(2006) - et al.
Tongue image analysis for appendicitis diagnosis
Information Sciences
(2005) - et al.
Towards automated enhancement, segmentation and classification of digital brain images using networks of networks
Information Sciences
(2001) - et al.
Color image watermark extraction based on support vector machines
Information Sciences
(2007) Spatial transformation models
- et al.
Classifying mental tasks based on features of higher-order statistics from EEG signals in brain-computer interface
Information Sciences
(2008) - et al.
Alzheimer’s diagnosis using eigenbrains and support vector machines
Electronics Letters
(2009) - et al.
Nonlinear spatial normalization using basis functions
Human Brain Mapping
(1999)
Neuropathological stageing of Alzheimer-related changes
Acta Neuropathologica
Analytic and iterative reconstruction algorithms in SPECT
The Journal of Nuclear Medicine
A tutorial on support vector machines for pattern recognition
Data Mining and Knowledge Discovery
An evaluation of maximum likelihood reconstruction for SPECT
IEEE Transactions on Medical Imaging
The diagnostic value of SPECT with tc 99 m HMPAO in Alzheimer’s disease. a population-based study
Neurology
Geometrical and statistical properties of systems of linear inequalities with applications in pattern recognition
IEEE Transactions on Electronic Computers
Statistical Parametric Mapping: The Analysis of Functional Brain Images
SVM feature selection for classification of SPECT images of Alzheimer’s disease using spatial information
Knowledge and Information Systems
Applications of support vector machines to speech recognition
IEEE Transactions on Signal Processing
Brain SPETC perfusion in early Alzheimer disease: where to look?
European Journal of Nuclear Medicine
An effective cluster-based model for robust speech detection and speech recognition in noisy environments
The Journal of the Acoustical Society of America
Textural features for image classification
IEEE Transactions on Systems, Man and Cybernetics
A comparison of classification methods for differentiating fronto-temporal dementia from Alzheimer’s disease using FDG-PET imaging
Statistics in Medicine
Cited by (0)
- ☆
This work was partly supported by the MICINN under the PETRI DENCLASES (PET2006-0253), TEC2008-02113, NAPOLEON (TEC2007-68030-C02-01) and HD2008-0029 projects and the Consejería de Innovación, Ciencia y Empresa (Junta de Andalucía, Spain) under the Excellence Project (TIC-02566).