Application of Empirical Mode Decomposition (EMD) on DaTSCAN SPECT images to explore Parkinson Disease

Abstract

Parkinsonism is the second most common neurodegenerative disorder. It includes several pathologies with similar symptoms, what makes the diagnosis really difficult. I-ioflupane allows to obtain in vivo images of the brain that can be used to assist the PS diagnosis and provides a way to improve its accuracy. In this paper a new method for brain SPECT image feature extraction is shown. This novel Computer Aided Diagnosis (CAD) system is based on the Empirical Mode Decomposition (EMD), which decomposes any non-linear and non-stationary time series into a small number of oscillatory Intrinsic Mode Functions (IMF) a monotonous Residuum. A 80-DaTSCAN image database from the “Virgen de las Nieves” Hospital in Granada (Spain) was used to evaluate this method, yielding up to 95% accuracy, which greatly improves the baseline Voxel-As-Feature (VAF) approach.

Introduction

Parkinsonian Syndrome (PS) or Parkinsonism is one of the most common neurodegenerative disorders, characterized by the presence of hypokinesia associated with rest tremor, rigidity or postural instability. Parkinsonism is commonly caused by Parkinson’s Disease (PD), a neurodegenerative disease which originates due to progressive loss of dopaminergic neurons of the nigrostriatal pathway. This cell loss involves a substantial decrease in the dopamine content of the striatum and a loss of dopamine transporters (Booij et al., 1997). Up to 2% population over 65 years suffer from PD and about 20–25% of all are not correctly diagnosed.

I-ioflupane (better known as DaTSCAN) is a neuroimaging radiopharmaceutical drug used to assist the diagnosis of Parkinsonism. This drug binds to the dopamine transporters in the striatum. Dopamine transporters (DAT) are proteins situated at the presynaptic terminal of dopaminergic neurons which are responsible for the re-uptake of dopamine. In order to obtain the distribution of the radiopharmaceutical in the brain and visualize the loss of dopamine, a Single-Proton Emission Computed Tomography (SPECT) camera is used (Booij et al., 1997, Winogrodzka et al., 2003). The SPECT maps obtained are normally examined by experts on PS diagnosis (Lozano, del Valle Torres, García, Cardo, & Arillo, 2007). These experts often use proprietary software to delimit Regions Of Interest (ROIs) and quantify the radiopharmaceutical uptake (Górriz et al., 2011, Jennings et al., 2004, Ortega Lozano et al., 2010, Pirker et al., 2000). This procedure is subjective and prone to errors since it relies on gross changes in transporter density throughout the ROI (Salas-Gonzalez et al., 2010). As such, it may not be sensitive to changes in the pattern of distribution that can characterize the progression of the disease (Scherfler & Nocker, 2009).

Recently some methods based on the machine learning paradigm have been applied to image analysis procedures, developing CAD systems for several neurodegenerative diseases (Seppi et al., 2006, Zubal et al., 2007), just like Alzheimer Disease or Parkinson Disease (Illán et al., 2012, Segovia et al., 2012). This CAD system is used for the extraction of complex high-dimensional features of the images that will be used to train the automatic classifier SVM to discriminate between normal and pathological controls (Illán et al., 2010, Illán et al., 2012, Ramírez et al., 2009), performing thus an automatic diagnosis (Chaves et al., 2012, Chaves et al., 2011, Chaves et al., 2012, Chaves et al., 2011, Martı´nez-Murcia et al., 2012, Ortiz et al., 2011).

The aim of this work is to continue the work undertaken by Gallix, Górriz, Ramírez, ÁIllán, and Lang (2012) and evaluate the performance of an adaptative image decomposition technique based on EMD (Zeiler et al., 2010). For this purpose we make use of different methods. Firstly we apply a preprocessing method consisting on an intensity normalization and a gaussian filtering of the images to remove noise added by the image reconstruction system in the Hospital. Once the noise is removed, images are decomposed with EMD and filtered again. Finally we make use of Principal Component Analysis (PCA) (Maćkiewicz and Ratajczak, 1993, Moore, 1981) and Independent Component Analysis (ICA) (Hyvärinen, 1999b) to extract their features and classify with a Support Vector Machines (SVM) method (Martínez et al., 2010, Segovia et al., 2012, Vapnik, 1998) as shown in Fig. 1.

This article is organized as follows. In Section 2 the image acquisition, preprocessing, adaptative image decomposition technique (EMD), feature extraction and classification methods used in this work are presented. In Section 3 we describe the experiment proposed, comparing the results obtained with the baseline results (VAF). Results are discussed in Section 3 too, and finally, conclusions are shown in Section 4.