International Islamic University in Malaysia

Problem Statement

Toddlers, unlike adults, are not able to tell their mothers how they feel and why they cry. It is difficult to determine If infants are experiencing pain because they have not yet developed language (Ranger et al., 2007). Existing research uses video recording of baby’s crying and facial expressions to detect whether or not a baby is suffering. But these phenomena can be observed for other factors, such as when a baby is hungry. Therefore, it is important to develop better methods for detecting and classifying infant pain using deep learning to process EEG signals from babies’ brain activities.

Research Objectives

1. To detect infant pain based on brain signals.
2. To classify infant pain using deep learning.
3. To implement an auto-detection and classification mechanism for infant pain

Research Questions

1. To which extent, the EEG Dataset can improve the ability to detect pain.
2. To which extent the implementation of deep EEG learning may improve the classification of pain in infants.

Operational Definitions

Electroencephalography

Electroencephalography is one of the most popular psychophysiological techniques in the clinic and preclinical.it provides a recording of electrical activity generated by brain structures

Pain

the International Association for the Study of Pain:” Pain is an uncomfortable sensual and emotional experience that is related to or described in terms of actual or potential tissue damage”.

Research hypothesis

During the infant pain, the magnitude of brain activity alters and EEG responses with different brain wave patterns.

Contribution

1. Use of EEG tool to collect samples.
2. Use deep learning to personalise infant pain and classify it.

Literature review

(Chang & Li, 2016)provide an automatic infant crying recognition method, in their work, the data were collected from National Taiwan University Hospital Yunlin Branch then by using Fast Fourier transform These data were translated into spectrogram which is used as an input to train the convolutional neuron networks (CNN). The CNN is adopted to train and classify the crying into three categories including hungry, pain and sleepy. The outcomes demonstrated that the trained CNN achieves high accuracy to classify the crying into the desired categories. The network achieves 78.5% validation accuracy, After running 25000 training iterations (Vatankhah & Toliyat, 2016) propose in their study the using of wavelet coherency as a pain level measurement tool in order to estimate the 3 levels of pain namely; no-pain, pain and unbearable pain. Besides, A Hidden Markov Model (HMM) and a support vector machine (SVM) techniques were used to classify these pain levels. The authors confirm by this study the hypothesis that the brain model under the mental task of chronic pain is mapped on EEG and the dependence of brain patterns on EEG is possible and detectable. SVM and HMM achieve the accuracy of (95%, 90%) and (78% ,65% )respectively (Chang & Li, 2016)(Zamzmi, Goldgof, Kasturi, Sun, & Science, 2018)propose a novel method for newborn pain expression detection using Transfer learning. They extract features from the dataset using pre-trained Convolutional Neural Networks and four classification algorithms which are: Naive Bayes, Nearest Neighbors (kNN), Support Vector Machines (SVMs), and Random Forests (RF) to classify baby’s facial expression whether it comes from pain or not. The result of their experimental shows that Transfer learning were a faster and more practical option than training CNN the scratch. for pain classification yielded the best result with 92.71% accuracy. (Zamzmi et al., 2018) on their work, try to identify the reason why infants cry whether its because of hungry, pain, sleepiness, discomfort OR its because of pathological status (asphyxia, deaf, jaundice, premature condition and autism, etc.); to do so, they chose to analyse cry signals as it contains divers levels of information, wavelet packet based energy and non-linear entropies were used as method to improve classification. From the results of this experiment, it can be apparently that the method proposed in this study gives a suitable accuracy. (Hariharan et al., 2017)

[bookmark: _Hlk532400555](Littlewort & Lee, 2007) proposes an automatic system for pain recognition from facial expression by using machine learning approach as individual facial actions detector to differentiate expressions of real pain and fake pain, The experiment consists of recording video from 26 participants under three experimental conditions: baseline, pose pain ,and real pain, In the real pain condition, subjects experienced cold pressor pain by soak their arm in ice water. The system was able to differentiate faked from real pain. The accuracy 52% for distinguishing fake from real pain. (Littlewort & Lee, 2007)

Critiques

The existent studies:

• Applied machine learning and EEG to detect adults’ pain not infant pain.
• Only few works that applied deep learning and EEG but detect adult pain.
• For those who analysed Facial expression, the variability in facial expression during nonpainful episodes can mimic pain(Bellieni, 2012), and change the classification direction in the case of infant’s face with scratch.
• For those who analyse the cry recording, the cry can be also a sign of other cases, such as hungry, and anger, Therefor, it cannot be used as the sole indicator of pain. (Bellieni, 2012).
• Almost all studies did the binary pain classification.

Methodology

To achieve the objectives, three methods are proposed; empirical methodology, observation experiments methodology and build methodology

Empirical Methodology

Empirical research will be applied in our work by collecting a large amount of EEG data from infant, and these data will be devised into two parts: the more important part will be used to train our system and the second part will be use for testing, the system will do wrong at first and it will take time to explore the data to obtain the desire results.

Observation Experiments Methodology

Observation experiments research is conducted to measure the electrical activity in each baby’s brain during crying, using the EEG. Measuring the electrical activity of the brain is important because the magnitude of brain activity alters and EEG responses with different brain wave patterns. These patterns can help the system to classify infant pain. the experience will address some questions related to what has been learned from these experiences, how many iterations has done to get the desire results, how the system reduce the error of learning, and the accuracy of the classification.

Build Methodology

Build methodology is proposed in this work; because it leads to the implementation of our software system, the requirements of this method are:

• Design the software system:

the implementation of a flexible platform

• Choose an adequate programming language:

we have chosen python as a programming language. Because it is easy to understand due to its consistent syntax and the way it reflects human language and/or its mathematical counterparts.

• Consider testing all the time:

Our system will get inputs which are the EEG infant dataset and these inputs will be running and testing time to get the desire results (the right classification).

Data Analysis

Figure 1 Data analysis process

Parent permission

At the hospital, we have to ask parent if they accept their children to participate in this work, just before the test is carried out, if they accept then they have to give us their consents for having an EEG test for her /his child.

If they give their consents, but then they change their mind, they can withdraw their consents at any time.

Data Recording

During EEG, electrodes are attached to baby’s scalp with a paste-like substance, the electrodes record the electrical activity of baby’s brain. If infant has pain, it’s common to have changes in his normal pattern of brain waves, Recording the EEG during cry period may help to determine what kind of pain the baby is having or rule out other conditions (hungry, sleepiness, discomfort)

EEG Artifacts Eliminating

Once the EEG signals have been recorded, the EEG recording will include artifacts such as muscle and eye artifacts, our system will not provide the best results if these undesirable input components, therefore; This study uses a bandpass filter to remove noisy artifacts such as EOG, EMG from the input.

Feature Extraction

Once the step of filtering artifacts from EEG is done, the feature extraction step is realised for the EEG recording to select suitable parameters for classification;

This is an important step because the obtained result will be used for the signal classification. A well determined feature extraction mechanism is necessary to obtain accurate classification of signal; Considering that several techniques have been implemented in the EEG signal, this study adopts Principal Component Analysis to extract features from EEG.

Donate your essay and get 10$ for each one!

Upload your essay and after it checking you will get money in your bonus account.

donate essay

Classification of pain

With a set of features at hand, a deep learning method will principally be employed to classify the infant pain from EEG signal in this study, a part of the EEG signals is used to train the deep learning algorithm to detect possible cases of pain in the infant. The second part of EEG dataset will use for the testing phase.

Bibliography

1. Tinmaswala, M. A., Valinjker, S. K., Hegde, S., & Taware, P. (2015). Electroencephalographic abnormalities in first onset afebrile and complex febrile seizures and its association with type of seizures. Journal of Medical Science and Clinical Research, 3, 7073-7082.
2. Zamzmi, G., Pai, C. Y., Goldgof, D., Kasturi, R., Sun, Y., & Ashmeade, T. (2016). Machine-based multimodal pain assessment tool for infants: a review. arXiv preprint arXiv:1607.00331.
3. Ditterrich, T. G. (1997). Machine learning research: four current direction. Artificial Intelligence Magzine, 4, 97-136.
4. Walter, S., Gruss, S., Limbrecht-Ecklundt, K., Traue, H. C., Werner, P., Al-Hamadi, A., … & Andrade, A. O. (2014). Automatic pain quantification using autonomic parameters. Psychology & Neuroscience, 7(3), 363-380.
5. Fuhr, T., Reetz, H., & Wegener, C. (2015). Comparison of supervised-learning models for infant cry classification/vergleich von klassifikationsmodellen zur säuglingsschreianalyse. International Journal of Health Professions, 2(1), 4-15.
6. Jones, L., Laudiano-Dray, M. P., Whitehead, K., Verriotis, M., Meek, J., Fitzgerald, M., & Fabrizi, L. (2018). EEG, behavioural and physiological recordings following a painful procedure in human neonates. Scientific data, 5, 180248.
7. Hariharan, M., Sindhu, R., Vijean, V., Yazid, H., Nadarajaw, T., Yaacob, S., & Polat, K. (2018). Improved binary dragonfly optimization algorithm and wavelet packet based non-linear features for infant cry classification. Computer methods and programs in biomedicine, 155, 39-51.
8. Hartley, C., Duff, E. P., Green, G., Mellado, G. S., Worley, A., Rogers, R., & Slater, R. (2017). Nociceptive brain activity as a measure of analgesic efficacy in infants. Science translational medicine, 9(388), eaah6122.
9. Lopez-Martinez, D., & Picard, R. (2017, October). Multi-task neural networks for personalized pain recognition from physiological signals. In Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW), 2017 Seventh International Conference on (pp. 181-184). IEEE.
10. Zamzmi, G., Goldgof, D., Kasturi, R., & Sun, Y. (2018). Neonatal Pain Expression Recognition Using Transfer Learning. arXiv preprint arXiv:1807.01631.
11. Bellieni, C. V. (2012). Pain assessment in human fetus and infants. The AAPS journal, 14(3), 456-461.
12. Vatankhah, M., & Toliyat, A. (2016). Pain level measurement using discrete wavelet transform. International Journal of Engineering and Technology, 8(5), 380.
13. Lopez-Martinez, D., & Picard, R. (2017, October). Multi-task neural networks for personalized pain recognition from physiological signals. In Affective Computing and Intelligent Interaction Workshops and Demos (ACIIW), 2017 Seventh International Conference on (pp. 181-184). IEEE.
14. Chang, C. Y., & Li, J. J. (2016, May). Application of deep learning for recognizing infant cries. In Consumer Electronics-Taiwan (ICCE-TW), 2016 IEEE International Conference on (pp. 1-2). IEEE.
15. Amaral, J. N. (2011). About computing science research methodology.
16. Teplan, M. (2002). Fundamentals of EEG measurement. Measurement science review, 2(2), 1-11.
17. Ayash, E. M. M. (2014). Research Methodologies in Computer Science and Information Systems. Retrieved November, 28, 2014.
18. Jeyaraman, S., Muthusamy, H., Khairunizam, W., Jeyaraman, S., Nadarajaw, T., Yaacob, S., & Nisha, S. (2018). A review: survey on automatic infant cry analysis and classification. Health and Technology, 1-14.
19. Amaro-Camargo, E., & Reyes-García, C. A. (2007, August). Applying statistical vectors of acoustic characteristics for the automatic classification of infant cry. In International Conference on Intelligent Computing (pp. 1078-1085). Springer, Berlin, Heidelberg.
20. Messaoud, A., & Tadj, C. (2011, May). Analysis of acoustic features of infant cry for classification purposes. In Electrical and Computer Engineering (CCECE), 2011 24th Canadian Conference on (pp. 000089-000092). IEEE.
21. Barajas-Montiel, S. E., & Reyes-García, C. A. (2006). Fuzzy support vector machines for automatic infant cry recognition. In Intelligent Computing in Signal Processing and Pattern Recognition (pp. 876-881). Springer, Berlin, Heidelberg.
22. Hadjistavropoulos, H. D., Craig, K. D., Grunau, R. E., & Whitfield, M. F. (1997). Judging pain in infants: behavioural, contextual, and developmental determinants. Pain, 73(3), 319-324.
23. Parkhi, O. M., Vedaldi, A., & Zisserman, A. (2015, September). Deep face recognition. In BMVC (Vol. 1, No. 3, p. 6).
24. Kira, K., & Rendell, L. A. (1992). A practical approach to feature selection. In Machine Learning Proceedings 1992 (pp. 249-256).
25. Walker, S. M. (2017). Translational studies identify long-term impact of prior neonatal pain experience. Pain, 158, S29-S42.
26. André, M., Lamblin, M. D., d’Allest, A. M., Curzi-Dascalova, L., Moussalli-Salefranque, F. T. S. N. T., Vecchierini-Blineau, M. F., … & Plouin, P. (2010). Electroencephalography in premature and full-term infants. Developmental features and glossary. Neurophysiologie clinique/Clinical neurophysiology, 40(2), 59-124.
27. Shellhaas, R. A. (2012). Continuous electroencephalography monitoring in neonates. Current neurology and neuroscience reports, 12(4), 429-435.
28. Eriksson, M., Storm, H., Fremming, A., & Schollin, J. (2008). Skin conductance compared to a combined behavioural and physiological pain measure in newborn infants. Acta paediatrica, 97(1), 27-30.
29. Gormally, S., Barr, R. G., Wertheim, L., Alkawaf, R., Calinoiu, N., & Young, S. N. (2001). Contact and nutrient caregiving effects on newborn infant pain responses. Developmental medicine and child neurology, 43(1), 28-38.
30. Barr, R. G., Rotman, A., Yaremko, J., Leduc, D., & Francoeur, T. E. (1992). The crying of infants with colic: a controlled empirical description. Pediatrics, 90(1), 14-21.
31. Diers, M., Koeppe, C., Diesch, E., Stolle, A. M., Hölzl, R., Schiltenwolf, M., … & Flor, H. (2007). Central processing of acute muscle pain in chronic low back pain patients: an EEG mapping study. Journal of clinical neurophysiology, 24(1), 76-83.
32. Diers, M., Koeppe, C., Diesch, E., Stolle, A. M., Hölzl, R., Schiltenwolf, M., … & Flor, H. (2007). Central processing of acute muscle pain in chronic low back pain patients: an EEG mapping study. Journal of clinical neurophysiology, 24(1), 76-83.
33. Holsti, L., Grunau, R. E., Oberlander, T. F., & Osiovich, H. (2008). Is it painful or not?: Discriminant validity of the Behavioral Indicators of Infant Pain (BIIP) scale. The Clinical journal of pain, 24(1), 83.²
34. Slater, R., Cornelissen, L., Fabrizi, L., Patten, D., Yoxen, J., Worley, A., … & Fitzgerald, M. (2010). Oral sucrose as an analgesic drug for procedural pain in newborn infants: a randomised controlled trial. The Lancet, 376(9748), 1225-1232.
35. Littlewort, G. C., Bartlett, M. S., & Lee, K. (2007, November). Faces of pain: automated measurement of spontaneousallfacial expressions of genuine and posed pain. In Proceedings of the 9th international conference on Multimodal interfaces (pp. 15-21). ACM.
36. Chang, C. Y., & Li, J. J. (2016, May). Application of deep learning for recognizing infant cries. In Consumer Electronics-Taiwan (ICCE-TW), 2016 IEEE International Conference on (pp. 1-2). IEEE.

Did you like this example?

4369

182