INTRODUCTION:
Human beings are known as Social
animals because of their ability to communicate with one another. The
communication can be direct through speech or indirect through emotions.
Therefore, Emotions forms an important part of Human’s life as they directly
affect the decision making skills of a person. In the recent years much of the
work has been carried out in studying these emotions and their different models.
Paul Eckman1 described six basic emotions that
are universally recognized throughout human cultures: fear, disgust, anger,
surprise, happiness, and sadness. Later, author expanded this list to include a
number of other basic emotions including embarrassment, excitement, contempt,
shame, pride, satisfaction, and amusement. These emotions play an important
role in Human Machine Interaction (HMI) system and hence have become a latest
research trend.
There are several ways of recognising these emotions-
from speech signals or facial expressions2,3,4,5,6 to analysing the
physiological or bio signals7,8,9,10,11,12. The main disadvantage
with the facial expressions based recognition is that emotions can be faked
with facial expressions. On the other hand emotions leave an immediate impact
on physiological signals. Several studies have been conducted on most of the
physiological signals except Wrist pulse signals directly. WPS or radial
signals have been used for centuries in Ancient Indian Ayurveda, Traditional
Chinese Medicines and Greek medicines to detect any abnormalities in the human
body13. Any physiological change in the body changes the parameters
related to pulse signals such as overall shape, strength, rate, rhythm etc. which
can be detected by feeling the palpations at three close points identified as
‘Vata’, ‘Pita, ‘Kapha’ on the radial artery. In the recent years various
researchers are working on these WPS by considering different conditions like
intake of meals, diseases, and exercise. Narendera et al. has discussed the
impact of food intake on the pulse of a person14 while Syed absar
kazmi et al. discussed about the impact of four different states of physical
activity- jogging, laying, standing and sitting on wrist pulse signals using
PPG method15. Also Bhaskar Thakkar has done health diagnosis using
Wrist pulse signals16.
The whole setup of the proposed work is divided into
various steps among which the first step is Pulse Acquisition. There are
different methods of pulse acquisition such as using pressure sensors,
ultrasonic sensors or photoelectric sensors17. The next step is the
signal pre processing which includes base line filtering of acquired signals.
Afterwards features are extracted and their statistical significance is
analysed. Heng yu Ping et al.18 specifies the optimum features
required for differentiating different emotions. C. Godin et.al.19
compared different physiological signals and the impact of different emotions
on them. To study wrist pulse signals under different emotional states, invoking
emotions is a tedious task and is subjective in nature. Generally games,
pictures, audio-videos are used for this purpose. Guillame Channel20
used the game to detect boredom; engagement and anxiety while E. H. Jang et.
al.21 use different techniques including audio visuals to invoke
human emotions like boredom, pain and surprise. Lastly, classification is
performed on the features to recognise respective emotions. The main objective
of the present work is to acquire wrist pulse signals under two emotional
states, pre-process it, extracting the relevant features out of it in time domain
and further classifying them to study the impact of various emotions on it.
The overall paper is divided into four sections. The first
section includes the subjects that participated, methodology of acquiring and
pre-processing the signals. The second section involves features extraction.
The third section explains the validation of the features and classification.
The fourth section includes the conclusions and future work.
METHODOLOGY:
The
whole methodology for the experiment was self designed and various steps were taken
into consideration. The flowchart given in figure 1 shows the complete
methodology.
DATA
ACQUISITION:
The
first step is the Data acquisition that involves the subjects, setup, the conditions
and the tools being used for acquiring the signals.
Subjects:
There were 8 voluntary subjects participating in this
experiment of the age group 20-22 years (Male-5 and Female-3). The signals were
acquired from the radial artery of left hand wrist and at the same time the
right hand of all the participants was busy in playing game which is used to
invoke emotions during the experiment. Before the start of experiment, all the participants
were explained about the purpose and method of this experiment and after that
all have to give a consent prior the data collection starts to ensure all are
aware of steps carried out in this study.
Figure
1 Flowchart showing the research methodology
Materials:
During data acquisition, pulse pressure sensor is
placed on the wrist and NI DAQ card PCI-6251 is being used along with the
LABView simulated code to acquire the WPS at the sampling frequency of 250Hz as
shown in Figure 2. The acquired pulse signal is in milivolts (weak signal) and
is infected by noise therefore further needs the amplification and filtration. Additionally
audio or video songs, Tetris game and International Affective Picture System
(IAPS) 22,23,24,25 are the some of the ways used by researchers to
invoke emotions. However here the Tetris game was used to stimulus emotion in
human during this experiment as it is easy to control the various difficulty
levels in the game. Also it is well known and requires only one hand of the
user. Moreover Tetris doesn’t require much mental skills. And as per the idea
proposed by Anandhukrishnan26 how games affect the anxiety in
children, selecting a game for this experiment was an appropriate choice.
The
Tetris game shown in Figure 3(a) and 3(b) with easy and difficult level were used
to invoke two states of emotions- Boredom and Anxiety respectively. The
subjects were pre-informed about the game so that they are well-verse with the
rules of the game. Moreover, the level of the game used to invoke emotion is
varying from subject to subject. Subjects were advised to identify the level of
game they feel most easy (Boredom level) and quite difficult (Anxiety level). Foregoing
to data acquisition of signals under two emotional states Neutral signals were taken
to consider it as the neutral state of the subjects. Subsequently, the deviations
of Boredom and Anxiety signals were observed from the Neutral signal.
Fig.
2 Block Diagram showing the Signal Acquisition setup
Fig.
3(a) Tetris game Easy Level (Boredom)
3(b)
Tetris game Difficult Level (Anxiety)
PREPROCESSING:
In
signal processing, pre-processing is one of the important step which is
required to be considered to avoid the unwanted signals, power line
interferences and baseline wandering. The acquired signals are raw signals that
need to be processed before using them for further analysis. The raw signal for
both the emotional states of subject 1 and 2 is represented in Figure 4 and Figure
5 respectively. The signals are de noised using wavelet function ‘dB9’ and are
smoothed using band pass filter with the band pass frequency of 0.4-6 Hz27.
Then the signals are segmented from the offset and the outlier segments are
removed. The resulting signals so obtained are the cleaned signals that are used
for further processing. A visual inspection of Figure 4 and Figure 5 clearly
indicate the usefulness of further processing of these signals as it represents
that the information carried in both the raw signals of same subject is
different for different emotional states.
Fig.
4 LABView representation of Anxiety and Boredom signal of Subject1
Fig.
5 LAB View representation of Anxiety and Boredom signal of Subject2
FEATURES
EXTRACTION:
As
the signals are acquired, the next step is the extraction of the features in order
to characterize physiological activity for different emotional states. In this
study, the time domain or statistical features proposed by PICARD28 are
extracted and investigated. The time domain or statistical features extracted
from WPS were mean, variance, standard deviation, mean of the absolute value of
the first difference and mean of the absolute value of the second difference
giving rise to 5 features in total per subject under each emotional state-Neutral,
Boredom and Anxiety. Two more features i.e. mean of the absolute value of the
first difference and mean of the absolute value of the second difference were
extracted after normalisation of WPS. Table 1 shows the list of the features
considered for analysis.
Table
1: List of Features extracted
|
Features
|
Expression
|
|
Mean
|
|
|
Variance
|
|
|
Accuracy
|
ACC = 
|
|
Standard Deviation
|
|
|
Mean of the absolute value
of first difference
|
|
|
Mean of the absolute Value
of Second difference
|
|
|
Mean of the absolute Value
of First difference of Normalised signal
|
|
|
Mean of the absolute Value
of Second difference of Normalised signal
|
|
Where;
xi: signal; n: total length of the
signals; normalised signal (xni) = 
CLASSIFICATION:
After extracting features and grouping them into
different feature sets, the next step is to train a classifier for a particular
emotion based on the feature set created per subject. There are several machine
learning algorithms available for this process that is being used in the
previous works29,30,31,32,33. However, choosing a particular
classifier has always been a challenging task because of the individualistic
nature of the signals. In the proposed work, four classifiers named Support
Vector Machine (SVM), Random Forest, Naïve Bayes and MetaMulti class are being
used with 10 folds cross validation procedure on the training sets.
(i)
SVM:
One of the widely used
machine learning algorithms for supervised classification is Support Vector
Machine (SVM). It is mostly preferred for classification as well as regression
applications. However, in research related to bio signal processing it is
used for classifying different classes/states. In this
algorithm, each data point is mapped as a point in n-dimensional feature space
(where n is number of features you have) with the value of each feature being
the value of a particular coordinate. Then, classification is carried out by
finding the hyper-plane that discriminates multiple classes very
competently.
(ii)
Naïve
Bayes:
Naïve Bayes classifiers is not an individual algorithm as it belongs to the family of
simple probability based
classifiers that implements Bayes’ theorem with well designed (naïve) independence presumptions between the features. Naïve
Bayes is a simple method of constructing classifiers by assigning class labels
to problems represented as vectors of feature values, where the class labels are
drawn from some finite set. The
key advantage of Naïve Bayes is that it necessitates a fewer number of training
samples to approximate the parameters essential for classification.
(iii)
Meta
Multi Class:
Multiclass
classification, as the name itself suggests, refers to classifying more than two classes such as classifying a
set of images of shapes which may be square, triangle, or circle. Multiclass
classification makes the presumption that each sample is marked to one and
solely one label, that is, a shape can either be a square or a triangle but not
both at the same instant.
(iv)
Random
Forest:
It is
a highly flexible, user friendly supervised machine learning algorithm that
gives considerable results even without hyper-parameter tuning at the majority
of instants. Because its simplicity, it is one of the commonly used algorithms.
Here multiple decision trees created that are merged together for accurate and
stable prognosis.
Hence,
it is necessary to demonstrate the capability of physiological signal like WPS
in depicting difference across various emotional states and is statistically
significant for further research. The subsequent section will demonstrate
effectiveness of the WPS in detecting different emotional states followed by
cross validation using 4 different classifiers.
RESULTS
AND DISCUSSION:
Extracted
Features
In
this section, outcomes of above discussed features and classification tools
have been discussed for discriminating boredom, anxiety and neutral states.
Also, p- values and t-test have been performed for testing statistical
significance of extracted features. In the beginning, total of 7 features have
been computed for different emotional states from 8 subjects. The Table 1-3 shows
the features extracted (total 7 statistical features) independently for neutral,
boredom and anxiety state.
Statistical
Significance of Features:
Since
the main aim of the work is to discriminate the signals taken under different
emotional states as a result the features extracted needs to be evaluated. A
paired t-test was performed and p-values were calculated for every feature of
all the subjects in all the above mentioned states. The null hypothesis statement for paired samples t-test is:
the average of the differences between the paired observations in the two
samples is zero. However, if the computed P-value is below 0.10, statistically
it signifies that the mean difference between the paired observations is
significantly different from 0. More the values away from 0.10, least
significant are those features in detecting emotions.
The
reference state is the neutral state of the subject taken at the start of
experiment as mentioned earlier.
Table
5: T-values and p-values of the Paired t test applied on features for boredom
conditions at significance level 0.10
|
Features
|
T-Value
|
p-Value
|
|
Mean
|
2.52
|
0.0004
|
|
Variance
|
0.57
|
0.58
|
|
Standard Deviation
|
2.11
|
0.07
|
|
Mean of Absolute value of
1st dif.
|
2.32
|
0.053
|
|
Mean of Absolute value of2nd
diff.
|
2.53
|
0.038
|
|
Mean of absolute of 1st
diff. of normalised signal
|
0.82
|
0.43
|
|
Mean of absolute of 2nd
diff. of normalised signal
|
1.15
|
0.28
|
Table
6: T-values and p-values of the Paired t test applied on features for anxiety
conditions
|
Features
|
T-Value
|
p-Value
|
|
Mean
|
1.24
|
0.25
|
|
Variance
|
0.95
|
0.37
|
|
Standard Deviation
|
1.07
|
0.35
|
|
Mean of Absolute value of
1st dif.
|
1.21
|
0.265
|
|
Mean of Absolute value of2nd
diff.
|
1.37
|
0.21
|
|
Mean of absolute of 1st
diff. Of normalised signal
|
2.1
|
0.07
|
|
Mean of absolute of 2nd
diff. Of normalised signal
|
3.35
|
0.01
|
Table
7: T-values and p-values of the Paired t test applied on features for both
boredom and anxiety conditions
|
Features
|
T-Value
|
p-Value
|
|
Mean
|
0.92
|
0.39
|
|
Variance
|
0.91
|
0.39
|
|
Standard Deviation
|
0.64
|
0.54
|
|
Mean of Absolute value of
1st dif.
|
0.74
|
0.47
|
|
Mean of Absolute value of2nd
diff.
|
0.87
|
0.40
|
|
Mean of absolute of 1st
diff. Of normalised signal
|
1.61
|
0.1
|
|
Mean of absolute of 2nd
diff. Of normalised signal
|
2.91
|
0.02
|
Table
4 gives the T-values and p-values of the Paired t-test applied on features for
boredom conditions (Boredom vs. Neutral). It has been observed that 4 out of 7
features were significant in this case. It shows the importance of the mean,
standard deviation, mean of the absolute value of first and second difference
features to distinguish the Boredom state from the Neutral state. Similarly, Table
5 and 6 gives the T-values and p-values of the Paired t-test applied on
features for anxiety conditions (Anxiety vs. Neutral) and Boredom V/s Anxiety
condition respectively. It was observed that for both the conditions mean of
the absolute value of first and second difference of the normalised signals
were found significant. The graphical representation of all significant features
clearly differentiates the different emotional states. Such graphical
representation for the condition Boredom V/s Anxiety is shown in Figure 6. From
the graph it is clearly visible that significant features are sufficient enough
to distinguish different emotional states.
Fig.
6 Representation of the significant stastical features studied under Boredom
v/s Anxiety
CLASSIFICATION:
At first, all features are taken into consideration during
classification of all the subjects for case1 (Boredom vs. Neutral), case2 (Anxiety
vs. Neutral) and case3 (Boredom vs. Anxiety cases). From the results shown in
Figure 5 for Boredom vs. Neutral (case1), it is clear that when all the
features were under consideration, the highest accuracy was obtained as 68.75% by
SVM followed by Random Forest, Naive Bayes and Meta Multi Class. The best
accuracy for the classification of Anxiety vs. Neutral (case2) has been
observed as 75% and again SVM gave better result than others followed by
MetaMulti Class, Random Forest and Naive Bayes for all the features shown in Figure
6. In the case3 (Boredom vs. Anxiety), the accuracy was achieved as 62.5% by
Meta Multi Class followed by SVM, Naive Bayes and Random Forest.
Thereafter,
only selected dominant features were used in each emotional condition (case1-3)
at the stage of classification and results are illustrated within Figure 7-9.
Fig. 7 Case1: Classification using all features and
significant features in Boredom state (Boredom vs. Neutral)
Fig. 8 Case2: Classification using all features and
significant features (Anxiety vs. Neutral)
Fig. 9 Case3: Classification of Boredom and Anxiety states considering all and
significant features
When only significant features were considered the
accuracies achieved were 68.75%, 75% and 62.5% given by SVM, SVM and Meta Multi
Class respectively in all the 3 cases. Consequently, it can be concluded that working
with only the dominant features is sufficient enough to distinguish three cases
effectively. Furthermore, it is noticed that the overall performance given by SVM
in classifying emotions is better than other classifiers while considering all
or dominant features except for case 3 in which Meta Multi Class is the best.
The other important observation that was observed from
this experiment is that till now as per the literature, EEG signals were the
only signals used to classify human emotions with a higher rate of accuracy
approximately 90%. However, WPS signals with an accuracy of 75% has given
sufficient proofs of its competency as well which can be enhanced further using
different techniques to acquire them or classifying them differently. This is
shown in figure 10.
Figure 10: Comparative analysis of WPS with EEG in
recognizing emotions
CONCLUSIONS:
This paper investigates statistical parameters for
classifying Boredom and Anxiety states using WPS. Efforts have been made to
illustrate the potential of WPS in analysing the emotions. WPS were taken for
three different states -Neutral (Reference), Boredom and Anxiety induced with the
help of Tetris game. The protocol for acquiring signals using game to invoke
emotion was designed. Furthermore, seven statistical features were extracted
and their significance was tested using p-values. Proficient classification
methods were used to validate the results. The study demonstrated that the most
considerable features are sufficient enough to discriminate the state of mind
(emotion) in all the above mentioned three cases. To our best knowledge till
now no research study have been used directly on the WPS for emotions. This
preliminary study achieved the satisfactory results using SVM classifier in all
the cases except in Boredom vs. Anxiety. Overall, this paper presents new
insight into the role of WPS in emotional state classification. Further in
future, more focus would be given to strengthen the database by increasing
subjects, features and emotion states (such as Happy, Sad, and Fear) to improve
the classification rate.
CONFLICT OF INTEREST:
The authors
declare no conflict of interest.
REFERENCES:
1.
Dalgleish
T and Power M. Handbook of Cognition and Emotion. John Wiley and Sons,
Sussex, UK. 1999.
2. Korkmaz Onur Erdem, AtasoyAyten. Emotion Recognition
from Speech Signal Using Mel- Frequency Cepstral Coefficients. 9th
International Conference on Electrical and Electronics Engineering (ELECO).
2015; 1254-1257.
3. Young-Mee Kim, Hye-Jeon Hong. Effect of Creative
Rhythmical Movement Program on Perceptional Motor Ability, Emotional Experience
and Facial Emotion Recognition Ability of Adults with Developmental Disability.
Research J. Pharm. and Tech. 2019; 12(3): 1031-1038.
4. Darekar R. V., Dhande, A.
P. Enhancing effectiveness of emotion detection by multimodal fusion of speech
parameters. International Conference on Electrical, Electronics, and Optimization
Techniques (ICEEOT). 2016; 3242-3246. IEEE
5. Matlovic Tomas, Gaspar Peter, Moro Robert, Simko Jakub,
Bielikova Maria. Emotions Detection Using Facial Expressions Recognition and
EEG. International Workshop on Semantic and Social Media Adaptation and Personalization
(SMAP). 2016;18-23.
6. De, A., Saha, A. A
comparative study on different approaches of real time human emotion
recognition based on facial expression detection. International Conference on Advances
in Computer Engineering and Applications. 2015;483-487.
7. Karthikeyan P., Murugappan M., Yaacob S. Analysis of
Stroop Color Word Test-Based Human Stress Detection using Electrocardiography
and Heart Rate Variability Signals. Arab J
Sci Eng. 2014; 39(3):1835-1847.
8.
Gong Ping, Ma H. T., Wang Yutong. Emotion
recognition based on the multiple physiological features. Proceedings of IEEE
international Conference on real time computing and robotics. 2016;140-143.
9.
Chen Jing, Hu Bin, Wang Yue, Dai Y.Q.,
Yao Yuan, Zhao S. A three stage decision framework for multi subjects emotions recognition
using physiological signals. IEEE International Conference on Bioinformatics
and Biomedicines. 2016;470-474.
10.
Verma G.K., Tiwary U.S. Multimodal
fusion framework: A multiresolutiion approach for emotion classification and
recognition from physiological signals. Journal of Neuroimage. 2013;102:162-172.
11.
AlZoubi Omar, D’Mello Sidney K.,
and Calvo Rafael A. Detecting Naturalistic Expressions of Nonbasic Affect Using
Physiological Signals. IEEE transactions on affective computing. 2012;3(3):298-310.
12.
Xin Chai,
Wang Qisong, Zhao Yongping, Liu Xin, Bai Ou, and Li Yongqiang. Unsupervised
domain adaptation techniques based on auto-encoder for non-stationary EEG-based
emotion recognition. Computers in biology and medicine.2016;79: 205-214.
13.
S. Upadhyaya. Nadi Vijyaya Ancient
Pulse Science. Chaukhamba Publishers, India, 2005.
14.
Khaire Narendra N., Joshi Yashwant
V. Diagnosis Of Diseases Using Wrist Pulse Signals For Classification Of Pre-Meal
And Post-Meal Samples. International Conference On Industrial Instrumentation
And Control(ICIC) College Of Engineering. 2015;866-869.
15.
Kazmi,
Syed Absar, Sheroz Khan, Othman O. Khalifa, and Mansoor Hussain Shah.
"Spectrum analysis of physiological signals of human activities."
In 2015 International Conference on Emerging Technologies (ICET),
pp. 1-6. IEEE, 2015.
16.
Thakker,
B., Vyas, A. L., Farooq, O., Mulvaney, D., and Datta, S. Wrist pulse signal
classification for health diagnosis. 4th International Conference on Biomedical
Engineering and Informatics (BMEI). 2011; 4: 1799-1805. IEEE.
17.
Zuo, W.,
Wang, P., and Zhang, D. Comparison of three different types of wrist pulse
signals by their physical meanings and diagnosis performance. IEEE journal
of biomedical and health informatics. 2014; 20(1): 119-127.
18.
Ping,
H. Y., Abdullah, L. N., Halin, A. A., and Sulaiman, P. S. A study of
physiological signals-based emotion recognition systems. International
Journal of Computers and Technology. 2013; 11(1): 2189-2196.
19.
Godin,
Christelle and Prost-Boucle, F and Campagne, Aurélie and Charbonnier, Sylvie and
Bonnet, Stéphane and Vidal, A. (2015). Selection of the most relevant physiological
features for classifying emotion. PhyCS 2015 - 2nd International Conference on
Physiological Computing Systems, Proceedings.
20.
Chanel,
G., Rebetez, C., Bétrancourt, M., and Pun, T. Emotion assessment from
physiological signals for adaptation of game difficulty. IEEE Transactions
on Systems, Man, and Cybernetics-Part A: Systems and Humans. 2011; 41(6): 1052-1063.
21.
Jang,
E. H., Park, B. J., Park, M. S., Kim, S. H., and Sohn, J. H. Analysis of physiological
signals for recognition of boredom, pain, and surprise emotions. Journal
of physiological anthropology. 2015;34(1): 25.
22.
Fleureau,
J., Guillotel, P., and Huynh-Thu, Q. Physiological-based affect event detector
for entertainment video applications. IEEE Transactions on Affective
Computing. 2015; 3(3): 379-385.
23.
Frantzidis,
C. A., Bratsas, C., Klados, M. A., Konstantinidis, E., Lithari, C. D., Vivas,
A. B. And Bamidis, P. D. (2010). On the classification of emotional biosignals
evoked while viewing affective pictures: an integrated data-mining-based
approach for healthcare applications. IEEE Transactions on Information
Technology in Biomedicine. 2010;14(2):309-318.
24. Agrafioti, F., Hatzinakos,
D., and Anderson, A. K. ECG pattern analysis for emotion detection. IEEE
Transactions on affective computing.2011; 3(1): 102-115.
25. Greco Alberto, Valenza Gaetano, Lanata Antonio, Rota Giuseppina,
and Scilingo Enzo Pasquale. Electrodermal Activity in Bipolar Patients during
Affective Elicitation. IEEE Journal of Biomedical and Health Informatics.
2014;18(6):1865-1873 .
26. Anandhukrishnan T. G, Anjali Sivan, Difna Baiju K,
Lavanya Raj R S, Sangeetha. S, Saritha S, Betcy K Jayims. Effectiveness of Play
on Anxiety among Hospitalized Children. Asian J. Nursing Education and
Research. 2018; 8(2): 225-229.
27.
Garg N., Ryait H.S., Kumar A., Bisht
A. An effective method to identify various factors for denoising wrist pulse
signal using wavelet denoising algorithm. Biomedical Materials and Engineering.
2017; 29(1): 53-65.
28.
Picard,
R. W., Vyzas, E., and Healey, J. Toward machine emotional intelligence:
Analysis of affective physiological state. IEEE Transactions on Pattern
Analysis and Machine Intelligence.2001; 10: 1175-1191.
29. Wen Wanhui, Liu Guangyuan, Cheng Nanpu, Wei Jie, Shangguan
Pengchao, and Huang Wenjin. Emotion Recognition Based on Multi-Variant Correlation
of Physiological Signals. IEEE Transactions On Affective Computing. 2014;
5(2):126-140.
30. Berta Riccardo, Bellotti Francesco, Gloria Alessandro
De, Danu Pranantha, and Schatten Carlotta. Electroencephalogram and
Physiological Signal Analysis for Assessing Flow in Games. IEEE Transactions on
Computational Intelligence and Ai in Games. 2014.5(2): 164-175.
31. Li Qing, Yang Zongkai, Liu Sanya, Dai Zhicheng, and Liu
Yang. The Study of Emotion Recognition from Physiological Signals. International
Conference on Advanced Computational Intelligence. 2015;378-382.
32. Jang E.H., Park B.J., Kim S.H., Sohn J.H. Classification of Three Different Emotion by
Physiological Parameters. Journal of the Ergonomics Society of Korea 2012;31(2):
271-279.
33. A. Mohamed Syed Ali . ECG signal classification based
on temporal and spectral features using SVM classifier . Research J. Pharm. and
Tech 2017; 10(11): 4116-4118.