2016 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)
978-1-5090-1610-5/16/$31.00 ©2016 IEEE 489
ELM-Based Classification of ADHD Patients Using a
Novel Local Feature Extraction Method
Yang Li, Zhichao Lian, Min Li, Zhonggeng Liu, Liang Xiao, Zhihui Wei
School of Computer Science and Engineering , Nanjing University of Science and Technology
Nanjing,China
Abstract—Recently, it has been an increasing interest in
modeling abnormal temporal dynamics of functional interactions
in psychiatric disorders. However, the accuracy of differentiating
attention-deficit/hyperactivity disorder (ADHD) children form
normal children has still much space for improvement. To further
improve the accuracy, the key issue is to extract more effective
features from original fMRI data. In this paper, we propose a
novel local feature extraction method named Local Binary
Encoding Method (LBEM) that can effectively characterize
functional interaction patterns (FIPs). In particular, we show that
the proposed method can well discriminate the functional
interaction abnormalities, which is composed of a Bayesian
connectivity change point model, a local feature extraction
method and a kernel Extreme Learning Machine (ELM) -based
classifier. The experiment on a real dataset of 23 ADHD children
and 45 normal control (NC) children has shown that our method
achieved better classification performance compared to the
existing methods.
Keywords—local feature extraction method, functional
interaction patterns, Bayesian connectivity change point model,
ELM, ADHD
I. INTRODUCTION
Modeling functional connectivity between anatomically
distinct regions of interest (ROIs) in the brain has emerged as
a powerful tool for investigating functional brain interactions
and their abnormalities in psychiatric conditions in recent years
[1]. In this work, we have applied a model based on novel
Bayesian connectivity change point model (BCCPM) [2] that
contains a local feature extraction method to analyze the
abnormal temporal dynamics of functional interactions
especially attention-deficit/hyperactivity disorder (ADHD).
Increasing the accuracy of differentiating ADHD children form
normal children on the basis of [1] is our pursuit.
Recent studies suggested that the functional
activity/connectivity of any cortical area is subject to top-down
influences of attention, expectation, and perceptual tasks [3].
Dynamic interactions between connections from higher to
lower-order cortical areas and intrinsic cortical circuits involve
moment-by-moment functional changes in brain [3]. It has
been reported that functional interactions are still undergoing
sizable dynamic changes at different time scales even in resting
state [4]. Along this study direction, there have been several
studies analyzing brain functional dynamics. In [5], M.A.
Lindquist et al. have developed Hierarchical Exponentially
Weighted Moving Average (HEWMA) method on fMRI
signals to detect BOLD signal state change in response to
stimulus. In [6], C. Chang et al. use sliding time windows to
capture the functional brain dynamics.
In this work, a novel classification framework composed of
an effective Bayesian connectivity change point model for
modeling functional brain interactions, a local binary encoding
method for extracting features and a kernel ELM based
classifier for classification is proposed for classifying ADHD
patients. The main novelty of the paper is a new local feature
extraction method named Local Binary Encoding. The
procedure includes three stages (see Fig. 1). First, the BCCPM
is introduced to split sample set for obtaining a better
understanding of the functional brain dynamics without any
prior knowledge. Second, a local binary encoding method is
exploited on each sample to extract more effective and
discriminable local features for robust classification. Finally, a
KELM classifier is applied for the processed brain data to
compute the final classification results.
Input split Extraction KELM Output
Fig. 1. The flowchart of the proposed computational framework.
II. METHOD
A. Bayesian Connectivity Change Point Model
As mentioned before, brain functional interaction
/connectivity is under dynamic changes. Traditional analyses
which extract features directly from the whole time series do
not consider the dynamics of functional connectivity, and thus
the classification results are not accurate. For the purpose of
detecting the dynamics of functional interaction patterns, we
need to segment the fMRI time series into blocks where the
functional interaction patterns are considered as static. In this
paper, we applied the Bayesian connectivity change point
model (BCCPM) proposed by Lian et al. [2] to detect the
boundaries segmenting time series into blocks representing
different functional interaction patterns. Here, we just briefly
introduced the BCCPM model and the details can be referred
to [2].
Given a data matrix Y that consists of m of variables (i.e.,
ROIs) and T observations in temporal order, the BCCPM can
detect the location of change points which segment the time
series into blocks with different joint probabilities (defining
functional interactions) among the m variables.
First, the BCCPM defines a block indicator vector
where
if the time t is a change point (defined as the
starting point of a temporal block), and
otherwise. The