Feature Extraction and Classification of NIRS Data

Feature Extraction
and Classification of
NIRS Data.
May 3, 2017
Dept. of ECE ,
KUET.
Presented By:
Pritam Mondal
Roll: 1209006
ECE 4000
Supervisor:
Dr. Sheikh Md. Rabiul Islam,
Associate Professor,

Outlines
• Objectives
• Introduction
• Basics of NIRS And Data Acquisition
• Proposed Methodology
• Simulation and Result Analysis
• Future Work
• Conclusions
• References
May 3, 2017 2

Objectives
 To extract different features from the NIRS data.
 To classify the extracted features.
 To find out the comparison among different parameters
of the classification.
May 3, 2017 3

Introduction
 The Near Infrared Spectroscopy (NIRS) is dependent on
changes of blood flow, as it measures oxygenated and
deoxygenated hemoglobin’s ([HbO] and [HbR]).
 The NIRS data is analyzed using different feature
extraction method and classification method.
 Principal Component , Independent Component
Analysis, Non Linear Principal Component , Median,
Mean are the features that are extracted.
May 3, 2017 4

Introduction (Cont.)
 Artificial Neural Network (ANN), k-nearest Neighbor
Algorithm (k-NN), Support Vector Machine (SVM) are the
methods of classification.
 Comparative review of different methods of features and
classification are also analyzed.
May 3, 2017 5

Basics of NIRS And Data
Acquisition
 Near-infrared spectroscopy (NIRS) is a spectroscopic
method that uses the near-infrared region of the
electromagnetic spectrum (from about 700 nm to
2500 nm).
 NIRS employs low-energy optical radiation (mostly in 2-
3 different wavelengths) to assess absorption changes in
the underlying brain tissue.
 These absorption changes reflect the changes in local
concentration of oxy- and deoxy-hemoglobin.
May 3, 2017 6

Acquisition (Cont.)
 NIRS can offer simultaneous measurements from
dynamic changes of Oxy-hemoglobin (HbO) and Deoxy-
hemoglobin (HbR) in the brain cortex.
 Changes in HbO is used to represent neural activity in
the human brain.
 Optical topography makes use the different
absorption spectra of oxygenated and deoxygenated
hemoglobin in the near infrared region.
May 3, 2017 7

Acquisition(Cont.)
 NIRS system produces 695 and 830 nm NIR signals
through frequency-modulated laser diodes
 This NIR signal is absorbed by hemoglobin, while the
non-absorbed NIR signals are reflected to the source
 Since Oxyhemoglobin (HbO) and Deoxyhemoglobin
(HbR) absorb NIR light differently, 2 wavelengths of
light (695 and 830 nm) are used
May 3, 2017 8

Acquisition (Cont.)
May 3, 2017
Figure 1: Channel configuration, right hemisphere (channels 1-12) and left hemisphere
(channels 13-24)
9

Proposed Methodology
May 3, 2017 10
NIRS Data
Feature
Extraction
• Mean
• Median
• PCA
• ICA
• NLPCA
Classification
• ANN
• k-NN
• SVM
Figure 2: Block diagram of the proposed methodology

Simulation and Result
Analysis (Cont.)
May 3, 2017 11
Figure 3: HbO2 (Oxy-Hemoglobin) data for all 24 channel for patient 1
right arm
1) Raw Data

Analysis
12May 3, 2017
Figure 4: HbR (Deoxy-Hemoglobin) data for all 24 channel
for patient 1 right arm

Analysis (Cont.)
May 3, 2017 13
Figure 5: Total (HbO2 -HbR) data for all 24 channel for
patient 1 right arm

Analysis (Cont.)
May 3, 2017 14
Figure 6: Mean and median for 24 channel (of total data for person 2
right arm)
2. Features

Analysis (Cont.)
May 3, 2017 15
Figure 7: Independent Component for 24 channel (of total data for person
2 right arm)

Analysis (Cont.)
May 3, 2017 16
Figure 8: Non Linear Principal Component for 24 channel (of total
data for person 2 right arm)

Analysis (Cont.)
3.Classification
May 3, 2017 17
Figure 9: Confusion matrix(ANN) for Non Linear Principal
Component (of total data for person 2 right arm)

Analysis (Cont.)
SL.
NO.
Feature
Extracti
on
Method
Classification
ANN k-NN SVM
Accuracy Sensitivity Accuracy Sensitivity Accuracy Sensitivity
1 Mean 70.8% 100% 100% 100% 37.5% 100%
2 Median 45.8% 50% 100% 100% 37.5% 100%
3 PCA 50% 50% 91.6% 100% 37.5% 100%
4 ICA 50% 50% 91.6% 100% 62.5% 100%
5 NLPCA 83.3% 100% 100% 100% 37.5% 100%
May 3, 2017 18
Table 1: Comparison of different parameter of classification of
different feature for patient 2 right arm(total)

Analysis (Cont.)
May 3, 2017 19
70.8
45.8
50 50
83.3
0
10
20
30
40
50
60
70
80
90
Mean Median PCA ICA NLPCA
Accuracy(%)
Feature
Figure 10: Comparison of accuracy of ANN for different features (of
Total data for person 2 right arm)

Analysis (Cont.)
May 3, 2017 20
100 100
91.6 91.6
100
86
88
90
92
94
96
98
100
102
Accuracy(%)
Feature
Figure 11: Comparison of accuracy of k-NN for different features (of

Analysis (Cont.)
May 3, 2017 21
37.5 37.5 37.5
62.5
37.5
0
10
20
30
40
50
60
70
Accuracy(%)
Feature
Figure 11: Comparison of accuracy of SVM for different features (of

Analysis (Cont.)
May 3, 2017 22
4. Result Analysis
 Some unique feature is found on ANN
 Accuracy is high for k-NN among ANN, k-NN and
SVM
 The sensitivity is also highest for NLPCA in most
cases (for ANN) than the other features
 It is found that for ANN NLPCA has the best
accuracy in almost all cases.

Future Work
 The extracted feature will be used for Brain Computer
Interface (BCI).
 Person authentication or mood detection using the NIRS
data will be tried also.
May 3, 2017 23

Conclusions
 Principal Component, Independent Component, Non
Linear Principal Component, Median, Mean are
extracted as features.
 By Artificial Neural Network (ANN), k-nearest Neighbor
Algorithm (k-NN), Support Vector Machine (SVM) the
classification is done.
 Some decisions are taken using the specifications of the
classification
May 3, 2017 24

References
• [1].M. Hatakenaka, I. Miyai, M. Mihara, S. Sakoda, and K. Kubota, “Frontal regions involved in learning of motor
skill—a functional NIRS study,” Neuroimage, vol. 34, pp. 109-116, 2007.
• [2].Yan Jiang, Xin Jin, Xin Li ,“Features Extraction from NIRS Data using Extreme Decomposition,” JOURNAL OF
COMPUTERS, VOL. 9, NO. 7, JULY 2014.
• [3].Nandish . M, Stafford Michahial, Hemanth Kumar P, Faizan Ahmed,“Feature Extraction and Classification of
EEG Signal Using Neural Network Based Techniques,” International Journal of Engineering and Innovative
Technology (IJEIT) Vol. 2, Issue 4, October 2012.
• [4]. R. Fernandez Rojas, X. Huang, K. L. Ou, D. Tran, and S. M. R. Islam, “Analysis of pain hemodynamic
response using near infrared spectroscopy (NIRS),” Int. J. Mult. Appl., vol. 7, pp. 31-42, 2015.
• [5]. M. Verner, M. J. Herrmann, S. J. Troche, C. M. Roebers, and T. H. Rammsayer, “Cortical oxygen
consumption in mental arithmetic as a function of task difficulty: a near-infrared spectroscopy approach,”
Front. Hum. Neurosci, vol. 7, 2013.
• [6]. Ramaswamy Palaniappan, “Biological Signal Analysis”.
• [7]. Bang-hua Yang, Guo-zheng Yan, Rong-guo Yan and Ting Wu, “Feature extraction for EEG-based brain–
computer interfaces by wavelet packet best basis decomposition,” JOURNAL OF NEURAL ENGINEERING, 2006.
• [8]. Tarik Al-ani and Dalila Trad, “Signal Processing and Classification Approaches for Brain-computer Interface,”
Intelligent and Biosensors, 2010.
• [10]. M. Kobayashi, Y. Otsuka, E. Nakato, S. Kanazawa, M. K. Yamaguchi, and R. Kakigi, “Do infants represent
the face in a viewpoint-invariant manner? Neural adaptation study as measured by near-infrared spectroscopy,”
Front. Hum. Neurosci. vol.5, 2011.
• [11].Y. Honda, E. Nakato, Y. Otsuka, S. Kanazawa, S. Kojima, M. K. Yamaguchi, et al., “How do infants perceive
scrambled face?: A near-infrared spectroscopic study,” Brain Res., vol. 1308, pp. 137-146, 2010.
May 3, 2017 25

References
• [12]. J. Gervain, F. Macagno, S. Cogoi, M. Peña, and J. Mehler, “The neonate brain detects speech structure,”
Proc. Natl. Acad. Sci. USA, vol. 105, pp. 14222-14227, 2008.
• [13]. I. Kovelman, M. H. Shalinsky, M. S. Berens, and L. A. Petitto, “Shining new light on the brain’s
“bilingual signature”: a functional Near Infrared Spectroscopy investigation of semantic processing,”
Neuroimage, vol. 39, pp. 1457-1471, 2008.
• [14]. M. Bartocci, L. L. Bergqvist, H. Lagercrantz, and K. Anand, “Pain activates cortical areas in the preterm
newborn brain,” Pain, vol. 122, pp. 109-117, 2006.
• [15]. C.H. Lee, T. Sugiyama, A. Kataoka, A. Kudo, F. Fujino, Y. W. Chen, et al., “Analysis for distinctive
activation patterns of pain and itchy in the human brain cortex measured using near infrared spectroscopy
(NIRS),” PLoS One, vol. 8, pp. e75360, 2013.
• [16]. S. M. Coyle, T. E. Ward, and C. M. Markham, “Brain-computer interface using a simplified functional near-
infrared spectroscopy system,” J. Neural Eng., vol. 4, pp. 219, 2007.
• [17]. M. Boecker, M. M. Buecheler, M. L. Schroeter, and S. Gauggel, “Prefrontal brain activation during stop-signal
response inhibition: An event-related functional near-infrared spectroscopy study,”Behav. Brain Res., vol. 176, pp.
259-266, 2007.
• [18]. X. Song, B. W. Pogue, S. Jiang, M. M. Doyley, H. Dehghani, T. D. Tosteson, et al., “Automated region
detection based on the contrast-to-noise ratio in near-infrared tomography,” Appl. Opt., vol. 43, pp. 1053-
1062, 2004.
• [19]. M. Verner, M. J. Herrmann, S. J. Troche, C. M. Roebers, and T. H. Rammsayer, “Cortical oxygen
consumption in mental arithmetic as a function of task difficulty: a near-infrared spectroscopy approach,”
Front. Hum. Neurosci, vol. 7, 2013.
• [20]. R. K. Hofbauer, P. Rainville, G. H. Duncan, and M. C. Bushnell, “Cortical representation of the sensory
dimension of pain,” J. Neurophysiol., vol. 86, pp. 402-411, 2001.
May 3, 2017 26

May 3, 2017
Thanks To All
Any Questions ?
27

Feature Extraction and Classification of NIRS Data

More Related Content

What's hot (20)

Similar to Feature Extraction and Classification of NIRS Data (20)

Recently uploaded (20)

Feature Extraction and Classification of NIRS Data