OPTIMAL POLYNOMIAL ELM FOR DETECTION OF NSI FROM SALIVARY SERS SPECTRA
TABLE OF CONTENT
HYPERLINK l _Toc4686 ACKNOWLEDGEMENT1
HYPERLINK l _Toc12334 ABSTRACT1
HYPERLINK l _Toc12459 LIST OF FIGURE1
HYPERLINK l _Toc29134 LIST OF TABLES1
HYPERLINK l _Toc12102 CHAPTER 1: INTRODUCTION1
HYPERLINK l _Toc23083 1.1 Introduction1
HYPERLINK l _Toc15809 1.2 Problem Statement1
HYPERLINK l _Toc31645 1.3 Research Objectives1
HYPERLINK l _Toc2019 1.4 Research Scope1
HYPERLINK l _Toc21102 1.5 Organization of thesis1
HYPERLINK l _Toc8677 CHAPTER 2: LITERATURE REVIEW1 HYPERLINK l _Toc15758
2.1 Detection of dengue disease3
HYPERLINK l _Toc15758 2.1.1 Long-Range Surface Plasmon-Polariton Waveguide Biosensers for Disease detection3
HYPERLINK l _Toc15758 2.1.2 Image Processing for detection of dengue virus based on WBC classification and decision tree3
HYPERLINK l _Toc15758 2.1.3 Classification o f salivary based NS1 from Raman Spectroscopy with support vector machine3
HYPERLINK l _Toc15758 2.1.4 Crystalization structure of whole saliva of drop coating deposition Raman for surface enhanced Raman spectroscopy analysis3
HYPERLINK l _Toc15758 2.1.5 Bio-functionalized tapered multimode fiber coated with dengue virus NS1 glycoprotein for label free detection of anti-dengue virus NS1 IgG antibody3
HYPERLINK l _Toc15758 2.2 Disease detection by using Raman Spectroscopy / SERS3
HYPERLINK l _Toc15758 2.2.1 Surface-enhanced Raman spectral analysis of substrates for salivary based disease detection3
HYPERLINK l _Toc15758 2.2.2 Non-invasive in vivo Raman spectroscopy of the skin for diabetes sceening3
HYPERLINK l _Toc15758 2.2.3 Point-of-care diagnosis of urinary tract infection (UTI) using surface enhanced Raman spectroscopy (SERS)3
HYPERLINK l _Toc15758 2.2.4 Linear Discriminant analysis for detection of salivary NS1 from SERS spectra3
HYPERLINK l _Toc15758 2.2.5 Automatic non-structural protein 1 recognition based on LDA classifier3
HYPERLINK l _Toc15758 2.2.6 Surface enhanced Raman spectrum of saliva for detection of lung cancer3
HYPERLINK l _Toc15758 2.2.7 Preliminary study in early detection technology of lung cancer based on surface-enhanced Raman spectroscopy3
HYPERLINK l _Toc15758 2.2.8 Applying principle component analysis for detection skin damage caused by using detergents: A Raman spectroscopy study3
HYPERLINK l _Toc15758 2.2.9 Noninvasive breast tumors detection based on saliva protein surface enhanced Raman spectroscopy and regularized multinomial regression 2.3 Disease detection by using ELM classifier3
HYPERLINK l _Toc15758 2.3 Disease detection by using ELM classifier3
HYPERLINK l _Toc15758 2.3.1 ICGA-ELM classifier for Alzheimer’s disease detection3
HYPERLINK l _Toc15758 2.3.2 Sleep apnoea episodes recognition by a committee of ELM classifiers from ECG signal3
HYPERLINK l _Toc15758 2.3.3 Detection of onset of Alzheimer’s disease from MRI image using a GA-ELM-PSO classifier3
HYPERLINK l _Toc15758 2.3.4 Macular Edema Severity detection in colour fundus images based on ELM classifier3
HYPERLINK l _Toc15758 2.3.5 Classification of Normal, ALS, and Myopathy EMG signals using ELM classifier3
HYPERLINK l _Toc15758 2.3.6 Sleep spindle detection using modified extreme learning machine generalized Radial Basis Function method3
HYPERLINK l _Toc28330 CHAPTER 3: THEORY3
HYPERLINK l _Toc15758 3.1 Raman Spectroscopy/ Surface Enhanced Raman Scattering (SERS)3
HYPERLINK l _Toc15758 3.2 Non Structural Protein 1 (NS1)3
HYPERLINK l _Toc15758 3.3 Principle Component Analysis (PCA)3
HYPERLINK l _Toc15758 3.4 Extreme Learning Machine (ELM)3
HYPERLINK l _Toc15758 3.4.1 ELM Algorithm3
HYPERLINK l _Toc15676 CHAPTER 4: METHODOLOGY3
HYPERLINK l _Toc5462 4.1 Procedural Flowchart3
HYPERLINK l _Toc29450 4.2 Extreme Learning Machine Flowchart3
HYPERLINK l _Toc9618 CHAPTER 5: RESULT & DISCUSSION3
HYPERLINK l _Toc29179 5.1 Scree Polynomial ELM3
HYPERLINK l _Toc15758 5.1.1 Case 1- Fixed parameter 1, vary parameter 23
HYPERLINK l _Toc15758 5.1.2 Case 2-Vary parameter 1, fixed parameter 23
HYPERLINK l _Toc19438 5.2 CPV Polynomial ELM3
HYPERLINK l _Toc15758 5.2.1 Case 1- Fixed parameter 1, vary parameter 23
HYPERLINK l _Toc15758 5.2.2 Case 2-Vary parameter 1, fixed parameter 23
HYPERLINK l _Toc32512 5.3 EOC Polynomial ELM3
HYPERLINK l _Toc15758 5.3.1 Case 1- Fixed parameter 1, vary parameter 23
HYPERLINK l _Toc15758 5.3.2 Case 2-Vary parameter 1, fixed parameter 23
HYPERLINK l _Toc26557 5.4 Comparative study3
HYPERLINK l _Toc6554 CHAPTER 6: CONCLUSION3
HYPERLINK l _Toc6554 REFERENCES3
ACKNOWLEDGEMENTFirst and foremost, I would like to thank Allah S.W.T, the Most Merciful and the Most Kind, for giving me the strength and enabling me to complete this Final Year Project 1 (FYP1) in the designated time period.
I would also like to thank University Teknologi Mara (UiTM) and thankful to Faculty of Electrical Engineering UiTM Shah Alam for giving me this opportunity to do my Final Year Project 1 in this semester. This 14 weeks of study period was a great learning curve under the kind guidance and support of my supervisors, Assoc. Prof. Dr. Lee Yoot Khuan. I would like to express special gratitude to them for their support and attention during this research period. I learnt a lot of new things from my supervisors and friends here. Next, thanks to my family for give such a big support in term of financial and moral to me to finish this Final Year Project 1 Research. Not to forget special thanks to NurShahirah binti Samsul Kamal, my friend because she always share the information with me and discuss the research project together.
Non Structural Protein 1 (NS1) is one of the biomarker for early recognition of dengue infection. Previous attempts to detect the dengue disease is using blood cell to detect the dengue. Blood cells are difficult to take as a test. There should be trained staff and special equipment to take blood samples. NS1 can be found in human saliva. Therefore using NS1 to detect dengue is an easy and safe way. Using Raman Spectroscopy or SERS, it can save time to detect the disease. If by the old method it is possible, it will take a long time to analyze the sample. The objective of our study is to evaluate the performance of ELM classifier. By using the ELM classifier, it can determine the accuracy of each sample that has been tested using SERS. Based on the results for the performance of the classifier used with the PCA criteria such as Scree, CPV and EOC can know which criteria is achieve with the desired characteristics. From the result, the best performance of PCA (Scree) accuracy, precision, sensitivity and specificity is 97.37%, 95%, 100% and 94.74%. Meanwhile, the best performance of PCA (CPV) accuracy, precision, sensitivity and specificity is 100%, 100%, 100% and 100%. And the best performance of PCA (EOC) accuracy, precision, sensitivity and specificity is 100%, 100%, 100% and 100%.
LIST OF FIGURES
HYPERLINK l _Toc14382 Figure 1.1: Dengue in Asia5
HYPERLINK l _Toc14382 Figure 1.2: Dengue cases in Malaysia5
HYPERLINK l _Toc14382 Figure 1.3: Symptoms of Dengue fever5
HYPERLINK l _Toc14382 Figure 1.4: Immune response to dengue infection5
HYPERLINK l _Toc14382 Figure 1.5: Raman Scattering5
HYPERLINK l _Toc14382 Figure 3.1: Raman Spectroscopy component5
HYPERLINK l _Toc14382 Figure 3.2: Nonstructural Protein 1 (NS1)5
HYPERLINK l _Toc14382 Figure 4.1: General Procedural Flowchart5
HYPERLINK l _Toc14382 Figure 4.2: ELM Flowchart5
HYPERLINK l _Toc14382 Figure 5.1: The accuracy, precision, sensitivity and specificity scree test (case 1)5
HYPERLINK l _Toc14382 Figure 5.2: The accuracy, precision, sensitivity and specificity scree test (case 2)5
HYPERLINK l _Toc14382 Figure 5.3: The accuracy, precision, sensitivity and specificity CPV test (case 1)5
HYPERLINK l _Toc14382 Figure 5.4: The accuracy, precision, sensitivity and specificity CPV test (case 2)5
HYPERLINK l _Toc14382 Figure 5.5: The accuracy, precision, sensitivity and specificity EOC test (case 1)5
HYPERLINK l _Toc14382 Figure 5.6: The accuracy, precision, sensitivity and specificity EOC test (case 2)5
LIST OF TABLES
HYPERLINK l _Toc14382 Table 2.1: detection of dengue disease5
HYPERLINK l _Toc14382 Table 2.2: Disease detection by using Raman Spectroscopy / SERS5
HYPERLINK l _Toc14382 Table 2.3: Disease detection by using ELM classifier5
HYPERLINK l _Toc14382 Table 5.1: Summary of result scree case 15
HYPERLINK l _Toc14382 Table 5.2: Summary of result scree case 25
HYPERLINK l _Toc14382 Table 5.3: Summary of result CPV case 15
HYPERLINK l _Toc14382 Table 5.4: Summary of result CPV case 25
HYPERLINK l _Toc14382 Table 5.5: Summary of result EOC case 15
HYPERLINK l _Toc14382 Table 5.6: Summary of result EOC case 25
Dengue disease is fast emerging pandemic-prone viral disease in many parts of the world. there are four types of dengue virus: DEN 1, DEN 2, DEN 3 and DEN 4. Within 3-14 days after infection bites, the symptoms will appear (average in 4-7 days). In urban poor areas, outskirts and inland areas are dying for dengue but also the richer neighborhoods in tropical and subtropical countries are also affected. Dengue is a viral infection caused by a female Aedes mosquito bite that can cause severe illness such as fever, headache, rash, muscle and joint pain, nausea and vomiting and sometimes cause potentially fatal complications. Severe dengue (previously known as dengue haemorrhagic fever) was first recognized in the 1950s during dengue epidemics in the Philippines and Thailand. Today it affects Asian and Latin American countries and has become a leading cause of hospitalization and death among children and adults in these regions.
Figure 1.1: Dengue in Asia
Figure 1.1 shows that dengue in Asia. The highest country with dengue is Philippines. A total of 166 thousand cases have been reported in the country. Malaysia is the fifth country with the most number of dengue diseases. A total of 43 thousand cases were reported in Malaysia. The lowest country has dengue disease in Asia is Singapore, as many as 22 thousand cases.
Figure 1.2: Dengue cases in Malaysia
Figure above shown that dengue cases by state in Malaysia. There are details all the state in the Malaysia about cases dengue. The highest dengue cases in 2017 are in Selangor. There are 29 thousand cases of dengue and there are 17 deaths caused by dengue. There were 28 cases and no deaths due to dengue in WP Labuan. WP Labuan is the lowest in dengue cases in 2017.
Figure 1.3: Symptoms of Dengue fever
Figure 1.3 shows symptoms of dengue fever. There are symptoms that can be known to detect dengue easily. If symptoms occur in the diagram, go to the nearest clinic to check for health. This dengue fever can cause death if it is not treated immediately.
Nonstructural Protein 1 (NS1)
Dengue virus (DENV) is a single-stranded RNA virus that belongs to the genus Flavivirus, which includes Zika virus, West Nile virus and Japanese Encephalitis virus. Dengue virus infection is transmitted to humans by mosquitoes, primarily Aedes aegypti species. Symptoms range from mild fever, to incapacitating high fever, with severe headache, pain behind the eyes, muscle and joint pain, and rash. Severe dengue is a potentially lethal complication that is characterized by fever, abdominal pain, persistent vomiting, bleeding and breathing difficulty.
Figure 1.4: Immune response to dengue infection
An infected person experiences the acute symptoms of dengue when there is a high level of the virus in the bloodstream. As the immune response fights the dengue infection, the person’s B cells begin producing IgM and IgG antibodies that are released in the blood and lymph fluid, where they recognize and neutralize the dengue virus and viral molecules such as the dengue NS1 protein. The immune response eliminates the virus, leading to recovery.
Raman Spectroscopy/ SERS
Raman scattering or the Raman effect is the HYPERLINK ;https://en.wikipedia.org/wiki/Inelastic_scattering; o ;Inelastic scattering; inelastic scattering of a HYPERLINK ;https://en.wikipedia.org/wiki/Photon; o ;Photon; photon by molecules which are excited to higher vibrational or rotational energy levels. When HYPERLINK ;https://en.wikipedia.org/wiki/Photons; o ;Photons; photons are HYPERLINK ;https://en.wikipedia.org/wiki/Scattering; o ;Scattering; scattered from an HYPERLINK ;https://en.wikipedia.org/wiki/Atom; o ;Atom; atomor HYPERLINK ;https://en.wikipedia.org/wiki/Molecule; o ;Molecule; molecule, most of them are HYPERLINK ;https://en.wikipedia.org/wiki/Elastic_scattering; o ;Elastic scattering; elastically scattered ( HYPERLINK ;https://en.wikipedia.org/wiki/Rayleigh_scattering; o ;Rayleigh scattering; Rayleigh scattering), such that the scattered photons have the same energy as the incident photons. A small fraction of the scattered photons are scattered inelastically by an excitation, with the scattered photons having a frequency and energy different from, and usually lower than, those of the incident photons. In a gas, Raman scattering can occur with a change in energy of a molecule due to a transition to another HYPERLINK ;https://en.wikipedia.org/wiki/Energy_level; o ;Energy level; energy level.
Figure 1.5: Raman Scattering
The different possibilities of light scattering: Rayleigh scattering (no exchange of energy: incident and scattered photons have the same energy), Stokes Raman scattering (atom or molecule absorbs energy: scattered photon has less energy than the incident photon) and anti-Stokes Raman scattering (atom or molecule loses energy: scattered photon has more energy than the incident photon).
Extreme Learning Machine (ELM)
Extreme learning machines are feedforward neural networks for classification, regression, clustering, sparse approximation, compression and feature learning with a single layer or multiple layers of hidden nodes, where the parameters of hidden nodes need not be tuned. In most cases, the output weights of hidden nodes are usually learned in a single step, which essentially amounts to learning a linear model. According to their creators, these models are able to produce good generalization performance and learn thousands of times faster than networks trained using back propagation. In literature, it also shows that these models can outperform support vector machines (SVM) and SVM provides sub optimal solutions in both classification and regression applications.
Firstly, from the literature survey, it found that the current method to detect the dengue is using the blood cell and certain antibody. However, there is no one except this group use saliva as a sample to detect dengue disease using Raman spectroscopy method and ELM classifier. Secondly, there is no research using Raman to detect dengue disease by using NS1 as a sample except this project. Most of them detect lung cancer, Diabetes and Urinary Tract Infection (UTI) using Raman. However, there is no one attempt to detect dengue by using SERS and NS1 as a biomarker. Lastly, from the literature review, ELM classifier has been used to classify diseases. Most using ELM classifier to detect disease related with brain. However, there is no attempt yet to classify dengue disease with NS1 using ELM classifier.
To develop a ELM classifier for detecting NS1 from salivary SERS spectra.
To optimize the ELM classifier using Polynomial for detection of dengue disease NS1 from salivary SERS spectra.
To evaluate the performance of Polynomial ELM for detection of dengue disease NS1 from salivary SERS spectra.
The scope of this project is to classify adulterated data with NS1 and without NS1 and compare the performance between the accuracy, precision, sensitivity and specificity. By using the dataset that uitm give is UITM-MMRR-12868-NS1-DENV. The data is loaded into MATLAB software to perform Polynomial ELM classifier to determine the accuracy, precision, sensitivity and specificity. There are three criteria of PCA, Cattell’s scree (Scree), Cumulative percentage of variance (CPV) and eigenvalues one citation (EOC). This research using polynomial so there are 2 parameter that can be vary. So for easily that can be classify to 2 case which is case 1 fixed parameter 1 and vary parameter 2. The case 2 is fixed parameter 2 and vary parameter 1. After that loaded the data into MATLAB software to perform the three criteria of PCA. Then compare the result follow the case and the criteria of the PCA.
ORGANIZATION OF THESIS
The thesis organization is the summarize for the entire chapters. Chapter 1 is about the details about introduction, problem statement, research objective and research scope for this project. Chapter 2 explanation about literature review of this project detection of dengue disease, disease detection by using Raman Spectroscopy / Surface Enhanced Raman Spectroscopy (SERS) and disease detection by using ELM classifier. Chapter 3 bring the theoretical of Raman Spectroscopy and Surface Enhanced Raman Spectroscopy (SERS), Non Structural protein1 (NS1), Principal Component Analysis (PCA) and Extreme Learning Machine. Chapter 4 describe background on data set and methodology in this study. Chapter 5 discusses the result on three kernel which is Scree, CPV and EOC. Chapter 6 is conclusion of this project.
2.1 DETECTION OF DENGUE DISEASE
2.1.1 LONG-RANGE SURFACE PLASMON-POLARITON WAVEGUIDE BIOSENSERS FOR DISEASE DETECTION
Long-range surface plasmon-polaritons (LRSPPs) are TM-polarized optical surface waves that propagate along a thin metal film or stripe in a homogenous cladding 1. The biosensors are constructed from metal stripe waveguides cladded in Cytop with etched microfluidic channels to expose the stripe surface to the sensing fluid. LRSPP waveguide biosensors are used to detection of leukemia, dengue fever, and urinary tract infection. LRSPP waveguide biosensers are used to detect immunoglobulin M (IgM) antibodies specifically in dengue disease in plasma blood samples of infected patients. In the human urine sample has a low concentration of constituents, so to selectively detect negative gram or gram positive bacteria in human urine samples by using LRSPP waveguide biosensors. The biosensor can detect bacteria and concentrations down to 105 CFU / ml, the internationally recommended threshold for the diagnosis of urinary tract infection 1. Using a functional strategy based on Protein G by using the LRSPP wave biosensor it is possible to detect the leukemia markers in the patient’s blood serum. The detection of leukemic abnormalities in serum was performed based on determining Formula concentration ratios Formula.
2.1.2 IMAGE PROCESSING FOR DETECTION OF DENGUE VIRUS BASED ON WBC CLASSIFICATION AND DECISION TREE
Dengue is caused by a virus transmitted from mosquitoes. Dengue is a major health problem in the tropics and the Asia-Pacific region, and it has spread faster than 50 years 2. Symptoms for such ailments such as fever, plasma leakage, etc. can lead to death. Risk reduction important factors of patients are getting diagnosed, and should be treated correctly and quickly.Previous research used image processing and image segmentation methods to classify white blood cells based on the edges of cells 2. The focus is selection of feature extraction and classification method. A white blood cells classification was done using image processing to apply the detection of dengue virus. The study proposed the models to classify white blood cells from blood cell image of 167 cells at accuracy of 92.2%. The models classified patients infected with dengue from 264 blood cell images at accuracy of 72.3% 2.
2.1.3 CLASSIFICATION OF SALIVARY BASED NS1 FROM RAMAN SPECTROSCOPY WITH SUPPORT VECTOR MACHINE
Non-structural protein 1 (NS1) is one of the eight nonstructural proteins encoded by the genome of Flavivirus genus. NS1 was first reported in 1970 as a viral antigen in the sera of dengue-infected patients 16. NS1 has been found to have an important role that is not yet clear for RNA replication. Our current study aims to classify NS1 infected saliva samples from healthy samples, a first ever attempt. NS1 saliva mixture samples were analyzed using SERS taken from saliva samples from healthy subjects. To learn the mode of vibration and rotation of the technique molecules that Raman Spectroscopy has used.It produces unique spectrum for each and every molecule from inelastic scattering of light17. The best performance is attained with RBF kernel with accuracy of 97.1% 93.4% 81.5% for 100ppm, 50ppm and 10ppm respectively 3.
2.1.4 CRYSTALIZATION STRUCTURE OF WHOLE SALIVA OF DROP COATING DEPOSITION RAMAN FOR SURFACE ENHANCED RAMAN SPECTROSCOPY ANALYSIS
Saliva is gaining attention as a diagnostic fluid in replacing blood for diagnosing systemic diseases. Most of the molecules present in blood can also be detected in salivary secretion 25. Therefore saliva indicates the health of the patient, in the event of a change in the patient’s saliva then the patient’s health is disrupted. Furthermore, it offers distinctive advantages over blood since it can be collected non-invasively without special training. Now, the usage ofsaliva as diagnostic fluid is expanding to systemic diseases including infectious diseases. For infectious diseases such as HIV26, dengue27, pigeon breeder’s disease (PBD) and lyme disease, specific antibodies are identified as the biomarker presents in saliva samples, where enzyme-linked immunosorbent assay (ELISA) is the common reported detection method used. Now with Surface Enhanced Raman Spectroscopy (SERS), which is capable of detection up to a single molecule, this problem can be surmounted 25. Saliva samples are easy to collect and can be done without trained staff and special equipment. If using a blood sample would have been practiced by trained staff and used special equipment. By using a saliva sample, it is safer and can prevent the risk of an infection of the bloodstream.
2.1.5 BIO-FUNCTIONALIZED TAPERED MULTIMODE FIBER COATED WITH DENGUE VIRUS NS1 GLYCOPROTEIN FOR LABEL FREE DETECTION OF ANTI-DENGUE VIRUS NS1 IgG ANTIBODY
Dengue fever has emerged as a polemic issue. Dengue infection has re-emerged as a major public health challenge worldwide. Reports of dengue disease infections have been increasing every year, it has been recorded over the last few years unreliable vaccine is visible.To demonstrates the development of a bio-functionalized tapered multimode optical fiber for the detection of anti-Dengue virus NS1 Immunoglobin G (IgG) 28. To detect Immunoglobin M in plasma blood by using long surface plasmon resonance also reported. However, methods require more from half an hour to detect the target of interest that can be better yet 29. Some advantages over other methods offered by alternative optical fiber-based biosensors. It has good sensitivity, low electromagnetic interference, easy to miniaturize, and applicable for in situ measurement with various applications in medical science field, biological engineering, food industry, and environmental monitoring 30.
Summary of literature review of detection of dengue disease is some of the previous research using blood and saliva to detect the dengue as medium. The biomarker that they using is NS1 and certain antibody. In previous research only 2 group are using SERS to detect the dengue disease but SVM classifier. This can conclude that, only our team are using the NS1 as biomarker, Raman spectroscopy or SERS as a method and ELM classifier to detect the dengue disease.
Year Title Author Objectives Disease Biomarker Method Medium Classifier Performance
2015 Long-Range Surface Plasmon-Polariton Waveguide Biosensors for Disease Detection Pierre Berini Application of long-range surface plasmon-polariton waveguide biosensors for disease detection leukemia, dengue fever, and urinary tract infection IGM antibody LRSPP waveguide Blood plasma N/A N/A
2015 Image Processing for Detection of Dengue Virus based on WBC Classification and Decision Tree Sarach Tantikitti, Sompong Tumswadi, Wichian Premchaiswadi Developed a method which can diagnose dengue fever disease. Dengue N/A Image Processing Red blood cell, White blood cell WBC and Decision Tree White blood cell
Red blood cell
2014 Classification of Salivary based NS1 from Raman Spectroscopy with Support Vector Machine A.R.M.Radzol, Khuan Y.Lee, W.Mansor To classify NS1 infected saliva samples from healthy samples, a first ever attempt. Dengue virus, West Nile Encephalitis virus, Yellow fever virus, Japanese Encephalitis virus NS1 SERS N/A SVM by using Linear, Polynomial and RBF kernal. Accuracy:97%
2012 Crystalization Structure of Whole Saliva of Drop Coating Deposition Raman for Surface Enhanced Raman Spectroscopy Analysis A.R.M.Radzol, Khuan Y.Lee, W.Mansor The characterization of dried whole saliva sample prepared via drop coating deposition Raman (DCDR) method for Raman analysis HIV, dengue, pigeon breeder’s disease and lyme disease Specific antibodies presents in saliva SERS Saliva N/A N/A
2018 Bio-functionalized Tapered Multimode Fiber Coated with Dengue Virus NS1 Glycoprotein for Label Free Detection of Anti-Dengue Virus NS1 IgG Antibody M.A. Mustapa, M.H.Abu Bakar, Y. Mustapha Kamil, Amir Syahir Amir Hamzah, M.A. Mahdi, Demonstrates the development of a bio-functionalized tapered multimode optical fiber for the detection of anti-Dengue virus NS1 IgG. Dengue IgG antibody bio-functionalized tapered optical fiber N/A N/A N/A
Table 2.1: detection of dengue disease
2.2 DISEASE DETECTION BY USING RAMAN SPECTROSCOPY / SERS
2.2.1 SURFACE-ENHANCED RAMAN SPECTRAL ANALYSIS OF SUBTRATES FOR SALIVARY BASED DISEASE DETECTION
Raman scattering is a light phenomenon observed by C.V.Raman and K.S Krishnan in 1928 31. This produces ‘molecular fingerprint’ for identifying molecules. However, the signal is extremely weak, causing Raman spectra not well received, not until the discovery of Surface Enhanced Raman Spectroscopy (SERS) 9. To study the structure of the solid molecular structure, the liquid and gas from its spread spectrum are Raman Spectroscopy. With this improvement, SERS has shown its niche in tracing molecular structure, especially in marking abnormal biological molecules such as cancer, conjunctivitis, AIDS 8. To be Raman reinforced to a variety that can be used should bind molecules to precious metal surfaces such as silver, gold and copper.From a biomedical perspective, the advantages of SERS including the number of samples for analysis is minimal, preparation for spectroscopy is minimal, analysis is easy and fast, the test is not damaging and easy to convert 8.
2.2.2 NON-INVASIVE IN VIVO RAMAN SPECTROSCOPY OF THE SKIN FOR DIABETES SCEENING
Diabetes affects more than 8.5% of the adult population; it is associated with over 3.7 million deaths per year 10.Raman spectroscopy is a promising method for non-invasive biomedical applications, such as screening for metabolic conditions. Keputusan ujian bagi menguji diabetes mengunakan Raman spectroscopy is accuracy 89.7%, sensitivity 89.2% dan specificity 92.0%. Furthermore, when the implemented ANN classifier showed a much higher accuracy than the minimal rate to assert statistical significance.
2.2.3 POINT-OF-CARE DIAGNOSIS OF URINARY TRACT INFECTION (UTI) USING SURFACE ENHANCED RAMAN SPECTROSCOPY (SERS)
Urinary tract infections are one of the most common types of infections causing millions of doctors visits and costs of billions of dollars every year in the US alone 12. Raman scattering, one of the whole organism fingerprinting techniques used, is observed when a very small number of photons incident on a molecule (about 1 in 107) are inelastically scattered11 .The weakness of the Raman’s effect resulting in a very low signal is the main limit of Raman, often below the detection limit for biological samples. SERS has been widely used in medicine. In this case, to identify samples as positive / negative for UTI and to get antibiotram against different antibiotics using SERS. By demonstrating a stronger approach to producing bacterial fingerprint with the minimum sample preparation using the Vibrational spectroscopic method. With this new method more effective. SERS spectra of serial dilutions of E. coli bacteria mixed with silver nanoparticles, showed a linear correlation between
spectral intensity and concentration 12.
2.2.4 LINEAR DISCRIMINANT ANALYSIS FOR DETECTION OF SALIVARY NS1 FROM SERS SPECTRA
Dengue virus (DENV) belongs to the Flavivirus genus of the Flaviviridae family 32. Biomarkers for early detection of DF are Protein non-structural 1 (NS1).The initial stage of dengue infection, especially in the first four days after the onset of symptoms, will be detected in infected blood and enzyme-related immunosorbent (ELISA) and immunechromatographic (ICT) tests used to detect NS1.Using ELISA, the presence of NS1 in saliva is detected, however with a low sensitivity performance. Raman Spectroscopy is suitable for the analysis of solid, liquid and gases samples. Surface Enhanced Raman spectroscopy (SERS) is an enhanced technique of Raman spectroscopy. It produces Raman spectra with higher intensity peak than the conventional Raman spectroscopy. Linear Discriminant Analysis (LDA) is a signal processing technique which can be used as a classifier 13. LDA is a widely-used classification method and it aims to find the optimum linear combination of the observed features to characterize or separate two or more classes of objects or events 33. It produces a new coordinate that yields maximal distance between the center points of the classes and minimal total variance within each class. The highest accuracy achieved is 98.4% with the corresponding sensitivity of 96.9%, precision of 100% and specificity of 100%.
2.2.5 AUTOMATIC NON-STRUCTURAL PROTEIN 1 RECOGNITION BASED ON LDA CLASSIFIER
NS1 is Non-structural protein type 1 found in a single strand of ribonucleic acid (RNA),of flavivirus genome in the family of flavivividae which cause fatal diseases including Japanese encephalitis (JE), Murray Valley encephalitis (MVE) , Tick-borne encephalitis, West Nile encephalitis (WNE), dengue fever (DF), and yellow fever 34. Within the first four days after the beginning of the NS1 symptom found to be distributed in blood serum of dengue virus. New markers for reported flavivirus infections are NS1 that can cause severe viral diseases such as encephalitis and dengue fever. NS1 protein also has the potential to distinguish bird flu from unusual human influenza viruses. Besides blood serum, NS1 is reported to be detected in saliva via enzyme-linked immunosorbent assay (ELISA) technique but with relatively low sensitivity and specificity.
2.2.6 SURFACE ENHANCED RAMAN SPECTRUM OF SALIVA FOR DETECTION OF LUNG CANCER
Lung cancer has the highest mortality rate because of the difficulty in early diagnosis. The diagnosis methods of lung cancer traditionally include physical examination, imaging tests, and biopsy 14. Lung cancer is ranked first in the mortality rate so the methods are inadequate for the detection rate of lung cancer. Saliva contains almost the same components as blood primarily excreted by sublingual glands, parotid glands and,submandibular glands, and ultrafiltration blood. Oral cancer, breast cancer, and pancreatic cancer using saliva for the diagnosis of these diseases. Mostly use enzyme-linked immunosorbent assay (ELISA), liquid chromatography, and mass spectrometry in routine biological techniques to analyze and it is very complicated and time-consuming. We use SERS for major investigations on saliva for the diagnosis of lung cancer. Surface enhanced Raman spectroscopy (SERS) can detect the bio-fluids at a molecular level, thus is capable of predicting lung cancer at an early stage. Principal component analysis combined with linear discriminant analysis was used for testing the diagnosing accuracy. We got a total accuracy of 82%.
2.2.7 PRELIMINARY STUDY IN EARLY DETECTION TECHNOLOGY OF LUNG CANCER BASED ON SURFACE-ENHANCED RAMAN SPECTROSCOPY
Lung cancer remains the leading cause of cancer mortality, and its incidence is increasing worldwide 35. The reason for the high mortality of lung cancer is 80% of lung cancer patients, when treatment is already in the late treatment of poor prognosis. To reduce effective early detection of lung cancer deaths is necessary.There has been no lung cancer screening capability to reduce the lung cancer death. Raman spectroscopy is a molecular scattering spectrum and is the effective means to study the molecular structure of substances. Human saliva contains abundant proteins and metabolites. Many diseases can be diagnosed by analysis of saliva 36. Using saliva for analysis makes it easier to sample collection and easier monitored, compared with sample of urine samples.Saliva test will be effective supplements from test available even replacements to other test methods, such as blood tests or urine 37. There is significant difference between lung cancer and normal human’s Raman spectra of saliva and the accuracy is up to 96.9% by Logistic Regression Analysis.
2.2.8 APPLYING PRINCIPLE COMPONENT ANALYSIS FOR DETECTION SKIN DAMAGE CAUSED BY USING DETERGENTS: A RAMAN SPECTROSCOPY STUDY
Raman Spectroscopy has been well established as apowerful non-invasive method for studying molecular structures and is a unique tool in medicine for non-invasive and real time analysis of biological tissues 38. Skin is a highly effective barrier to chemicals, and microbes. The skin can also control temperature, body water and electrolyte. An interesting finding was reported that in Basal Cell carcinoma, the most common cancer of skin, the collagen signal contribution in Raman spectrum is markedly reduced. In our daily life we are materials like chemical agents that are often used which can cause irritation and degeneration. Chemical agents are a way to detect and prevent skin disorders. To investigate alterations in molecular structures of skin by means of Raman Spectroscopy and Principle Component analysis.
2.2.9 NONINVASIVE BREAST TUMORS DETECTION BASED ON SALIVA PROTEIN SURFACE ENHANCED RAMAN SPECTROSCOPY AND REGULARIZED MULTINOMIAL REGRESSION
Breast cancer is the most common malignancy tumor and is the fifth leading cause of cancer death for women in China 39. To present a noninvasive breast tumors detection method using saliva protein surface enhanced Raman spectroscopy (SERS) and regularized multinomial regression (RMR) techniques through human saliva sample. Various types of proteins available in saliva can be used as a diagnosis of deadly human disease. In human biological samples there are many advantages in human saliva. This is because saliva is noninvasive and is very easy for storage and transportation. Besides that, saliva samples are also easily accessible and harmless. The study results showed that for RMR diagnostic model, the diagnostic accuracy of 92.78%,95.87% , and 88.66% are acquired, while discriminating among the normal group, the benign breast tumor group, and the malignant breast tumor group 39.
From the literature review of disease detection using Raman spectroscopy or SERS that can see at table 2.2 most of the previous research are using SERS to detect cancer. That no other group using the SERS to detect the dengue disease but only our group using the SERS to detect the dengue. They also using the saliva as their medium to detect the disease using Raman spectroscopy or SERS. The classifier that been use using SERS to detect cancer in previous research mostly used LDA classifier and no group that using ELM classifier. In our team group, there are using SERS to detect the dengue disease, using ELM classifier, NS1 as a biomarker and saliva as a medium.
Year Title Author Objectives Disease Biomarker Method Medium Classifier Performance
2012 Surface-Enhanced Raman Spectral Analysis of Substrates for Salivary based Disease Detection A.R.MRadzol, Y.K.Lee, W.Mansor, S.R.Yahaya To integration of the latest SERS in optical sensing technology and nanotechnology Cancer, conjuctivitis, AIDS N/A SERS Fingerprint N/A N/A
N/A Non-invasive in vivo Raman spectroscopy of the skin for diabetes screening Edgar Guevara, Juan Carlos Torres-Galvan, Miguel G.Ramirez Elias, Claudia Luevano-Contreras, Francisco Javier Gonzales To describes the application of portable Raman spectroscopy coupled with Artificial Neural Networks, to discern between diabetic patients and healthy controls Diabetes N/A N/A Skin Artificial neural network (ANN) Accuracy:89.7%
2012 Point-of-Care Diagnosis of Urinary Tract Infection (UTI) Using Surface Enhanced Raman Spectroscopy (SERS) Katerina Hadjigeorgiou,Evdokia Kastanos, Alexandros Kyriakides, Costas Pitris Using conventional methods, all three stages require bacterial cultures in order to provide results. UTI bacteria SERS Urinary N/A N/A
2017 Linear Discriminant Analysis for Detection of Salivary NS1 from SERS Spectra N.H.Othman, Khuan Y.Lee, A.R.M.Radzol, W.Mansor, N.N.M.Ramlan Using SERS detection of low concentration of NS1 in saliva seems promising Lung cancer, colorectal cancer, breast cancer, and oral cancer NS1 SERS with PCA N/A LDA Accuracy: 98.4%
2013 Automatic Non-Structural Protein 1 recognition based on LDA Classifier F.M.Twon Tawi, Khuan Y.Lee, W. Mansor, A.R.M.Radzol Raman spectra of saliva and NS1-saliva mixture are obtained using SERS N/A NS1 Raman Spectrum of Blank GS and Saliva, Raman Analysis, LDA Analysis N/A LDA N/A
2011 Surface Enhanced Raman spectrum of Saliva for detection of lung cancer Xiaozhou Li, Tianyue Yang, Rong Wang, Weidong Wen Principal components analysis combined with linear discriminant analysis was used for testing the diagnosing accuracy. Lung cancer N/A SERS Saliva PCA-LDA Accuracy: 82%
2010 Preliminary study on Early Detection Technology of Lung Cancer based on Surface-Enhanced Raman Spectroscopy Yan Wang, Shuang Sun, Dian Qu, Anyu Chen, Zijian Cui, Yulu Yao, Yi Jiao, Xun Guo, Chunwei Liu Analyzes the data of surface-enhanced Raman spectroscopy of the saliva which come from the issues of 19 lung cancers and 45 normal people. Lung cancer N/A SERS Saliva N/A Accuracy: 96.9%
2009 Applying Principle Component Analysis for Detecting Skin Damage Caused by Using Detergents: A Raman Spectroscopy Study Neda Baheri, Mohammad Hossein Miran Baygi, Rasoul Malekfar To investigate the damage a chemical detergent may have on skin structures and to determine the main vulnerable part of the skin. Skin Damage N/A Raman Spectroscopy Skin PCA N/A
2015 Noninvasive Breast Tumors Detection based on
Saliva Protein Surface Enhanced Raman
Spectroscopy and Regularized Multinomial
Regression Weilin Wu, Haiming Gong, Mingyu Liu, Guannan Chen, Rong Chen To present a noninvasive breast
tumors detection method using saliva protein surface enhanced
Raman spectroscopy (SERS) and regularized multinomial
regression (RMR) techniques through human saliva sample. Breast
tumors N/A SERS Saliva RMR Normal
Table 2.2: Disease detection by using Raman Spectroscopy / SERS
2.3 DISEASE DETECTION BY USING ELM CLASSIFIER
2.3.1 ICGA-ELM CLASSIFIER FOR ALZHEIMER’S DISEASE DETECTION
Alzheimer’s Disease (AD) is a progressive, neurodegenerative disorder that leads to memory loss, problems in learning, confusion and poor judgment 4. They require a significant computational effort for the training process and have a slower learning speed even though AD detection method uses SVM classifier and ANN model has been widely used.The high dimensional features with sample imbalance also influence their performance significantly 4. Comparison of ICGA-ELM and SVM Classifier performance comparisons are presented in this study to make a difference between the classifier. This comparison is made because the ICGA-ELM classifier reaches a better spread. ICGA-ELM produced a mean testing accuracy of 91.86% whereas, SVM produced a mean testing accuracy of 86.84%. ICGA-ELM classifier also produced a mean sensitivity of 0.91 and a mean specificity 0.94 whereas, SVM classifier produced a mean sensitivity of 0.89 and mean specificity 0.87 for 10 features. The accuracy of the ELM classifier testing is much higher than the SVM classifier.
2.3.2 SLEEP APNOEA EPISODES RECOGNITION BY A COMMITTEE OF ELM CLASSIFIERS FROM ECG SIGNAL
Epidemiologic studies have shown that obstructive sleep apnoea is a widespread medical condition and undiagnosed in a large number of adults 6. This sleep-related respiratory disorder is accompanied by interruptions in breathing during sleep and may lead to cardiovascular disease 7. To get higher performance need to be introduced new apnea identification tools that can be used more convenient sensors and implementing better signal in processing techniques. To improve accuracy it should implement better signal processing and machine learning techniques while using less sensors and signals that facilitate their application as homebased diagnostic tool.The maximum accuracy was 82.5% with a sensitivity of 81.9% and a specificity of 82.8%. The results were comparable to other published research with the same input data 5. The results indicate that the classification performance of the network committee is higher than the ELM classifier results.
2.3.3 DETECTION OF ONSET OF ALZHEIMER’S DISEASE FROM MRI IMAGE USING A GA-ELM-PSO CLASSIFIER
Alzheimer’s Disease (AD) is a progressive, neurodegenerative disorder afflicting humans. AD leads to memory loss, resulting in confusion, poor judgment and learning disabilities 40. Magnetic Resonance Imaging (MRI) scans are a new method for detecting the start of Alzheimer’s disease (AD). The three classes for classification considered in this study are normal, very mild AD and mild and moderate AD subjects. The GA-ELM-PSO classifier yields an average training accuracy of 94.57% and a testing accuracy of 87.23%, averaged across the three classes, over ten random trials.
2.3.4 MACULAR EDEMA SEVERITY DETECTION IN COLOUR FUNDUS IMAGES BASED ON ELM CLASSIFIER
Diabetic Macular Edema is a complication of diabetic retinopathy which is a cause of vision loss. It is a severe and widely spread eye disease 41. Swelling in the macular region is known as diabetic macular edema which is a cause of blindness 44. The automatic detection of hard exudates is useful in a complete analysis of retinal images. The detection performance has a sensitivity of 99% with specificity between 85% and 98%. The severity classification accuracy is 98% for the abnormal images.
2.3.5 CLASSIFICATION OF NORMAL, ALS, AND MYOPATHY EMG SIGNALS USING ELM CLASSIFIER
Muscles activities of human body are controlled by the electrical impulses generated by nervous system. Electromyography (EMG) is a technique of collecting and analyzing the signals at skeletal muscle 42. EMG signals are collected by placing electrodes at different places of muscles 45. Information on the status and function of muscles for diagnosis diseases such as Amyotrophic Lateral Sclerosis (ALS) and myopathy will be given by EMG signals. Myopathy involves stiffness of the muscles, and dysfunction while ALS is progressively neuromuscular, deadly disease.Using surface EMG signals three diagnosis marker with linear discrimination analysis classifier has been used for the diagnosis of the ALS diseases . Features such as area, duration, spike area, phase and turns have been employed with SVM and RBFNN classifier for the discrimination of normal, neuropathy and myopathy EMG signals 42.
2.3.6 SLEEP SPINDLE DETECTION USING MODIFIED EXTREME LEARNING MACHINE GENERALIZED RADIAL BASIS FUNCTION METHOD
Spindles are the main indicators of stage two sleep. Diseases such as Alzheimer’s and Schizophrenia are related to spindles reported by scientists. Many attempts have been made automatically for tracing spindles, and it’s boring and time-consuming. The Best average results of MELM-GRBF classifier after 15 training trials were, 93.10%, 90.34% and 95.90% for accuracy, sensitivity and specificity, respectively and the variances were 0.78, 1.61 and 1.16, respectively. While the ELM-RBF results for accuracy, sensitivity and specificity were 91.06%, 85.83% and 96.32% with 1.64, 2.39 and 1.18 variance, respectively 43.
Summary for literature review disease detection using ELM classifier is most the previous research using ELM classifier to detect the disease related with brain which is Alzheimer’s disease and sleep disorder. No one of them using Raman spectroscopy and SERS to detect the disease using ELM classifier. The medium that been use more to brain and the biomarker that the previous research using is blood cell. In our research, our team using ELM classifier to classify the performance of dengue disease by using SERS method and using NS1 as a biomarker.
Year Title Author Objectives Disease Biomarker Method Medium Performance
2013 ICGA-ELM Classifier for Alzheimer’s Disease Detection B.S.Mahannad, S.Suresh, N.Sundararajan, M.Aswatha Kumar To detect Alzheimer’s disease using voxel-based morphometric features and an extreme learning machine classifier Alzheimer’s disease N/A ICGA Brain ELM:
2015 Sleep Apnoea Episodes Recognition By A Committee of ELM Classifiers From ECG Signal Nadi Sadr, Philip de Chazal, Andre van Schaik, Paul Breen A committee of five ELM classifiers has been employed to classify one minute epochs of ECG into normal or apnoeic epochs Sleep apnoea diagnosis N/A ECG signal N/A Accuracy: 82.5%
2013 Detection of onset of Alzheimer’s Disease from MRI image using a GA-ELM-PSO Classifier Saraswath, B.S. Mahanand,
A.Kloczkowski, S. Suresh, N. Sundararajan. To get improved results and based on three classification problem Alzheimer’s disease N/A MRI Brain Normal
2017 Macular Edema Severity Detection in Colour
Fundus Images Based on ELM Classifier Jerald Jeba Kumar, Ravichandran C detection of hard exudates and
classification of diabetic macular edema severity from
color fundus images. Diabetic Macular Edema Blood
vessels in the retina image processing
techniques Eye Accuracy: 98%
2016 Classification of Normal, ALS, and Myopathy EMG
Signals Using ELM Classifier Vipin K Mishra, Varun Bajaj, Anil Kumar new technique based on five
features and Extreme learning machine (ELM) classifier for the
diagnosis of ALS and myopathy diseases. ALS and myopathy diseases N/A Electromyography (EMG) Muscles Accuracy: 88%
2014 Sleep Spindle Detection Using Modified Extreme Learning Machine Generalized Radial Basis Function Method Amin Hekmatmanesh, Seyed Mohammad Reza Noori, Mohammad Mikaili Designing algorithms to extract certain spindle features to diagnose them with high accuracy is valuable. Sleep Spindle N/A EEG signal N/A MELM-GRBF
Table 2.3: Disease detection by using ELM classifier
3.1 RAMAN SPECTROSCOPY / SURFACE ENHANCED RAMAN SCATTERING (SERS)
The Indian Physicist Sir Chandrasekhara Venkata Raman introduces the Raman Spectroscopy in 1928. The Raman Effect was name after the Indian Scientist Sir C.V.Raman. Raman won the Nobel Prize in Physics in 1930 18. Figure 3.2 shows the Raman Spectroscopy components.
Figure 3.1: Raman Spectroscopy component
To study the vibration and rotational modes of molecules used a technique Raman spectroscopy 21. Spectrum of photons scattered light can be pulled by Raman Spectroscopy experiencing frequency shift, due to the force of monochromatic light into the molecules. Every molecule produces uncommon spectrum from inflexible scattering of light. Raman scattering is a weak signal that can be disturb the Raman Spectroscopy practical application 21.
Raman spectroscopy has a wide variety of applications in biology and medicine. It has helped confirm the existence of low-frequency phonons19 in proteins and DNA, promoting studies of low-frequency collective motion in proteins and DNA and their biological functions 20.This is why Raman signal need high sensitivity analysis because of its effect are very weak. This is because of Raman signal are low efficiency, low concentration of samples and the sample is small quantities or volume.
Raman signal can be improve by using SERS markedly by substrates or molecules, which have absorbed by Nano-sized noble metal. This substrate is sensitive and selective which has an impact on the strength of the magnetic field is higher, as well as increase Raman scattering and amplify the intensity. For normal cases, the intensity can be increased up to 104-106 and for special cases could reach 108 and 1014 with the appropriate substrate 21.
In 1974, Fleichmann discovered Surface enhanced Raman spectroscopy (SERS). At that moment, they observed strong Raman enhancement on a electrochemically roughened silver electrode recoated 23.
Surface-enhanced Raman spectroscopy (SERS) is an extension of Raman spectroscopy, where metallic nanostructures are used to enhance the intensity of Raman scattering 22. SERS is a surface sensitive technique that improved the signal of Raman spectroscopy by amplify the molecules of Raman scattering. The molecules of the sample are absorbs into nano-size noble metal known as substrate. This absorption will have intensified Raman scattering then produce high peak spectrum. SERS is capable to detect up to single molecules. To fabricate the SERS substrate, need to use noble metal such as silver, gold or copper. There are many types of substrates such as metal roughened electrode, nanostructures, metal island films and colloids 24.
SERS is a new thing that can target living cells by using nanoparticles of gold and silver of light scattering that cannot be changed. To get excellent fabrication of SERS substrates use gold and silver. This is because of the surface Plasmon absorption is powerful in the detectable region of the spectra. Biochemical can also be analyzed in terms of limitless multiplexing and single molecule sensitivity. The technique of SERS detection sensitivity raise up to 1014 orders of magnitude.
3.2 NON STRUCTURAL PROTEIN 1 (NS1)
In 2006, NS1 was introduced for dengue testing. NS1 antigen test (nonstructural protein 1), is an antigen brought about by virus of the flaviviridae family, belonging to the flavivirus genome, which emcompasses single-stranded ribounucleic acid (RNA) virus 3. Dengue fever, shock syndrome, liver failure, dehydration, and encephalitis are severe illnesses in humans caused by Flavivirus Infection. NS1 can detect rapidly on the first day of the fever, before antibodies appear within a few days. NS1 protein is an antigen that has been recognized as a biomarker for diagnosis of flavivirus viral infections at the early stage .Through enzyme-linked immunosorbent assay is a method for detection of diseases. NS1 has been used to detect diseases around 40 countries. In human saliva, NS1 is easy to find. To detect dengue, using NS1 is easier and saves time. Saliva is gaining attention as a diagnostic fluid in replacing blood for diagnosing systemic diseases. In saliva secretion is also detected most of the molecules present in the same blood in the saliva. Recently, the high circulating levels of DENV-1 and DENV-2 NS1 in the acute phase of disease were demonstrated by use of the ELISA method 46. Therefore it is believed that the changes in saliva are indicative of wellness of patient.
Figure 3.2: Nonstructural Protein 1 (NS1)
3.3 PRINCIPAL COMPONENT ANALYSIS (PCA)
PCA was invented in 1901 by Karl Pearson,47 as an analogue of the principal axis theorem in mechanics, it was later independently developed and named by Harold Hotelling in the 1930s. PCA is a statistical procedure that uses an orthogonal transformation to convert a set of observations of possibly correlated variables into a set of values of linearly uncorrelated variables called principal components. This transformation is defined in such a way that the first principal component has the largest possible variance (that is, accounts for as much of the variability in the data as possible), and each succeeding component in turn has the highest variance possible under the constraint that it is orthogonal to the preceding components. The resulting vectorsclarification needed are an uncorrelated orthogonal basis set. PCA is sensitive to the relative scaling of the original variables.
PCA is mostly used as a tool in exploratory data analysis and for making predictive models. It’s often used to visualize genetic distance and relatedness between populations. PCA can be done by eigenvalue decomposition of a data covariance (or correlation) matrix or singular value decomposition of a data matrix, usually after mean centering (and normalizing or using Z-scores) the data matrix for each attribute.4 The results of a PCA are usually discussed in terms of component scores, sometimes called factor scores (the transformed variable values corresponding to a particular data point), and loadings (the weight by which each standardized original variable should be multiplied to get the component score).5
PCA is the simplest of the true eigenvector-based multivariate analyses. Often, its operation can be thought of as revealing the internal structure of the data in a way that best explains the variance in the data. If a multivariate dataset is visualised as a set of coordinates in a high-dimensional data space (1 axis per variable), PCA can supply the user with a lower-dimensional picture, a projection of this object when viewed from its most informative viewpoint. This is done by using only the first few principal components so that the dimensionality of the transformed data is reduced.
PCA is closely related to factor analysis. Factor analysis typically incorporates more domain specific assumptions about the underlying structure and solves eigenvectors of a slightly different matrix.
falseS2 = false
3.4 EXTREME LEARNING MACHINE (ELM)
Extreme learning machine (ELM) is a new learning algorithm for the single hidden layer feedforward neural networks. Compared with the conventional neural network learning algorithm it overcomes the slow training speed and over-fitting problems. ELM is based on empirical risk minimization theory and its learning process needs only a single iteration 48. .According to their creators, these models are able to produce good generalization performance and learn thousands of times faster than networks trained using back propagation. In literature, it also shows that these models can outperform support vector machines (SVM) and SVM provides sub optimal solutions in both classification and regression applications.
3.4.1 ELM ALGORITHM
Given a single hidden layer of ELM, supposed that the output of the false-th hidden nodes is false, where false and false are the parameters of the false-th hidden node. The output function of the ELM for SLFNs with L hidden nodes is:
false, where falseis the output weight of the false-th hidden node.
false is the hidden layer output mapping of ELM. Given N training samples, the hidden layer output matrix H of ELM is given as:
false and T is the training data target matrix: false
General speaking, ELM is a kind of regularization neural networks but with non-tuned hidden layer mappings (formed by either random hidden nodes, kernels or other implementations), its objective function is:
Minimize: false where false
Different combinations of falseand q can be used and result in different learning algorithms for regression, classification, sparse coding, compression, feature learning and clustering.
The example of kernel function are:
Radial Basis Function
The ELM classifier have been use to calculate the performance of the accuracy, precision, sensitivity, specificity and cohen’s kappa. There are the equation of the performance ELM classifier:
4.1 Procedural Flowchart
Figure 4.1: General Procedural Flowchart
From figure 4.1 shown that the general procedural flowchart. The flowchart is starting from collect the dataset from UITM. The dataset from UITM name is UiTM-MMRR-12868-NS1-DENV. The dataset is for load the training and testing dataset. After load the data set follow by the training and testing, it will divide into three criteria which is Scree, CPV and EOC for the training and testing dataset. After that normalize the dataset with ELM classifier. In this research use polynomial kernel. Polynomial equation is falsepower . The value of a and power need to vary to get the performance of classifier. After simulate the data using MATLAB software, it will come out with result for the performance. The performance of the classifier include the accuracy, precision, sensitivity, specificity and Cohen’s Kappa.
4.2 ELM Flowchart
Figure 4.2: ELM Flowchart
Figure 4.2 is ELM flowchart. Extreme learning machines are feedforward neural networks for classification, regression, clustering, sparse approximation, compression and feature learning with a single layer or multiple layers of hidden nodes, where the parameters of hidden nodes need not be tuned. In most cases, the output weights of hidden nodes are usually learned in a single step, which essentially amounts to learning a linear model. The performance of classifier based on accuracy, precision, sensitivity, specificity and Kappa. Principle Component Analysis (PCA) is the extraction of algorithm. There are three criteria of the PCA to be testing in this research. To detect the dengue using ELM classifier, there must calculate the performance of the classifier according to the criteria PCA that been use.
RESULT & DISCUSSION
5.1 Scree Polynomial ELM
The result will use 2 different case in order to get the exact result using the scree test which is the case 1 is fixed the parameter 1 and vary the parameter 2. For case 2 is fixed parameter 2 and vary parameter 1.The fixed value that been use in this research is 2 until 20 but the value of range that vary the parameter is from 0.0001 until 10000. In case 1 and case 2, the higher accuracy,precision, sensitivity and specificity is 97.37%, 95% , 100% and 94.74%.
5.1.1 Case 1- Fixed parameter 1, vary parameter 2
Table 5.1: Summary of result scree case 1
Figure 5.1: The accuracy, precision, sensitivity and specificity scree test (case 1)
Table 5.1 is shown the summary of scree test polynomial in case 1, fixed parameter 1 and vary parameter 2. In this table is the highest value for all parameter that been vary. The vary value is from 0.0001, 0.001, 0.01…. 10000. The fixed value is between 2 until 20. There are highest value for the accuracy, precision, sensitivity and specificity. The result can clearly if it in the graph like in figure 5.1. The graph is from the summary of result in table 5.1. At parameter 1= 2 & 3, the highest accuracy is starting when vary parameter 2 =1 but at parameter 1=4 and above, the highest accuracy start when parameter 2 vary at 0.0001. The highest accuracy, precision, sensitivity and specificity for this scree test case 1 is 97.37%, 95%, 100% and 94.74%.
5.1.2 Case 2-Vary parameter 1, fixed parameter 2
Table 5.2: Summary of result scree case 2
Figure 5.2: The accuracy, precision, sensitivity and specificity scree test (case 2)
Table 5.2 is shown the summary of scree test polynomial in case 2, fixed parameter 2 and vary parameter 1. The result can clearly if it in the graph like in figure 5.2. In the figure 5.2, the graph is not so smooth and not achieve the characteristic of graph for scree test. The lowest of the value accuracy is 92.11% and the highest is 97.37%. In scree test case 2, the graph did not constant at one value like the case 1 graph in figure 5.1. The highest value is different between the fixed and vary parameter. The highest accuracy, precision, sensitivity and specificity for this scree test case 2 is 97.37%, 95%, 100% and 94.74%.
Therefore, the highest accuracy, precision, sensitivity and specificity for scree test polynomial ELM is same for the both result but in the scree test for the polynomial ELM is the case 1 is better than case 2. It is because the graph for the case 1 is smooth and constant at highest accuracy value 97.37%.
5.2 CPV Polynomial ELM
The result CPV polynomial test also will use 2 different case in order to get the exact result which is the case 1 is fixed the parameter 1 and vary the parameter 2. For case 2 is fixed parameter 2 and vary parameter 1.The fixed value that been use in this research is 2 until 20 but the value of range that vary the parameter is from 0.0001 until 10000. The higher accuracy,precision, sensitivity and specificity is 100%, 100% , 100% and 100%.
5.2.1 Case 1- Fixed parameter 1, vary parameter 2
Table 5.3: Summary of result CPV case 1
Figure 5.3: The accuracy, precision, sensitivity and specificity CPV test (case 1)
For table 5.3 is the summary of CPV polynomial test for case 1. As you can see the graph in figure 5.3 is the accuracy, precision, sensitivity and specificity for the CPV case 1 test from the table 5.3. The value of the sensitivity is maintain constant 100% but the accuracy, precision and specificity will increase at certain point. The value for accuracy starting from 97.37% and increase to 100% at point parameter 1=13 and parameter 2=10. It also same with precision and specificity start from 95% and 94.74% then increase to 100%. So the CPV case 1 is achieve 100% accuracy, 100% precision, 100% sensitivity and 100% specificity.
5.2.2 Case 2-Vary parameter 1, fixed parameter 2
Table 5.4: Summary of result CPV case 2
Figure 5.4: The accuracy, precision, sensitivity and specificity CPV test (case 2)
For table 5.4 is the summary of CPV polynomial test for case 1. As you can see the graph in figure 5.4 is the accuracy, precision, sensitivity and specificity for the CPV case 2 test from the table 5.4. The result are maintain constant at 100% but with different of parameter that been vary. The performance is very smooth and achieved the characteristics of the graph. Parameter 2=2,3,…..20, the highest performance maintain 100%. So the CPV case 2 is achieve 100% accuracy, 100% precision, 100% sensitivity and 100% specificity.
Can conclude that from the result, the case 2 for CPV polynomial test is the best performance because it achieve 100% accuracy, 100% precision, 100% sensitivity and 100% specificity.
5.3 EOC Polynomial ELM
There are 2 different case in order to get the exact result using the EOC polynomial test which is the case 1 is fixed the parameter 1 and vary the parameter 2. For case 2 is fixed parameter 2 and vary parameter 1. In this test, the vary value is from 0.0001 until 10000 and the fixed value is from 2 until 20. The higher accuracy,precision, sensitivity and specificity is 100%, 100% , 100% and 100%.
5.3.1 Case 1- Fixed parameter 1, vary parameter 2
Table 5.5: Summary of result EOC case 1
Figure 5.5: The accuracy, precision, sensitivity and specificity EOC test (case 1)
For table 5.5 is the summary of EOC polynomial test for case 1. There are highest value for the accuracy, precision, sensitivity and specificity. The result can clearly if it in the graph like in figure 5.5. All value of the performance of accuracy, precision, sensitivity and specificity is maintain constant at 100% with different value of parameter 1 & 2. So the EOC case 1 is achieve 100% accuracy, 100% precision, 100% sensitivity and 100% specificity.
5.3.2 Case 2-Vary parameter 1, fixed parameter 2
Table 5.6: Summary of result EOC case 2
Figure 5.6: The accuracy, precision, sensitivity and specificity EOC test (case 2)
Table 5.6 is shown the summary of EOC test polynomial in case 2, fixed parameter 2 and vary parameter 1. As you can see the graph in figure 5.6 is the accuracy, precision, sensitivity and specificity for the EOC case 2 test from the table 5.6. The result can clearly if it in the graph like in figure 5.6. The result are maintain constant at 100% but with different of parameter that been vary. The performance is very smooth and achieved the characteristics of the graph. The highest accuracy, precision, sensitivity and specificity for this scree test case 2 is 100%, 100%, 100% and 100%.
Can conclude that from the result, with the both cases for EOC polynomial test, the both have same accuracy 100%, precision 100%, sensitivity 100% and specificity 100%. Both cases in EOC test is the best performance.
5.4 Comparative Study
Based on the 3 polynomial ELM test that is scree, CPV and EOC test, the CPV and EOC test achieve the 100% accuracy but the scree test only 97.37% the highest accuracy. If compare the CPV test and EOC test, the EOC test is better than CPV test because with both cases on the EOC test get the 100% accuracy and maintain constant. In CPV test the case 1 will achieve 100% accuracy at some point. So that can be conclude the EOC polynomial ELM test is the best performance than the scree and CPV.
The purpose of this work here is to develop a ELM classifier for detecting NS1 from salivary SERS spectra. By using the ELM classifier, that have classify the performance of the classifier. NS1 is used as a biomarker to detect the dengue disease. NS1 is easy to obtain because there is an NS1 protein in human saliva. The result shown that the ELM classifier is suitable to detect the dengue. NS1 also one of the biomarker that suitable and the result have been achieve. In this case, the polynomial have 2 parameter need to vary. Facilitate the division into 2 parts, case1 and case 2, for case 1 is fixed parameter 1 and vary parameter 2 and for case 2 is fixed parameter 2, var parameter 1. Result has shown criteria used and cases used to simulate the data.
It is found that, the optimization the ELM classifier using Polynomial for detection of dengue disease. The kernel used should be compatible with ELM classfier characteristic. For this activation function, one kernel need to be used which is polynomial kernel. There are three criteria of PCA, Cattell’s scree (Scree), Cumulative percentage of variance (CPV) and eigenvalues one citation (EOC) for simulate the result. The highest performance of accuracy for scree test is 97.37% but the highest performance of accuracy for CPV and EOC is 100%. From the three criteria, compare that which one is the better for the Polynomial ELM classifier.
From the result, it can be observed that the performance of Polynomial ELM for detection dengue disease achieve the characteristic. Based on the three criteria, there are the best performance among the three of criteria. The best performance is perform by the result and based on both cases. In this project, the best performance of ELM classifier is by using EOC criteria. It is because for the both cases in the EOC test is maintain constant at 100% the highest accuracy. So the EOC test is the best performance.
Pierre Berini, “Long-Range Surface Plasmon-Polariton Waveguide Biosensors for Disease Detection,”2015 International Topical Meeting on Microwave Photonics (MWP),2015, pp.1-2, 2015.
Sarach Tantikitti, Sompong Tumswadi, Wichian Premchaiswadi, “Image Processing for Detection of Dengue Virus based on WBC Classification and Decision Tree,” 2015 13th International Conference on ICT and Knowledge Engineering, 2015, pp.84-89, 2015.
A.R.M.Radzol, Khuan Y. Lee, W.Mansor, “Classification of Salivary based NS1 from Raman Spectroscopy with Support Vector Machine,”2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2014, pp.1835-1838, 2014.
B.S.Mahannand, S.Suresh, N.Sundararajan, M.Aswatha Kumar, “ICGA-ELM Classifier for Alzheimer’s Disease Detection,” 2013 Indian Conference on Medical Informatics and Telemedicine (ICMIT), 2013, pp.48-52, 2013.
Nadi Sadr, Philip de Chazal, Andre van Schaik, “Sleep Apnoea Episodes Recognition By a Committee of ELM classifiers from ECG signal,” 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2015, pp.7675-7678, 2015.
T. Young, P. E. Peppard, and D. J. Gottlieb, “Epidemiology of Obstructive Sleep Apnea,” Am. J. Respir. Crit. Care Med., vol. 165, no. 9, pp. 1217–1239, May 2002.
G. S. Hamilton, P. Solin, and M. T. Naughton, “Obstructive sleep apnoea and cardiovascular disease.,” Intern. Med. J., vol. 34, no. 7, pp. 420–6, Jul. 2004.
A.R.M.Radzol, Y.K.Lee, W.Mansor, S.R.Yahaya, “Surface-Enhanced Raman Spectral Analysis of Substrates for Salivary based Disease Detection,” 2012 IEEE 8th International Colloquium on Signal Processing and its Applications, 2012, pp505-509.
M. Fleischmann, P. J. Hendra, and A. J. McQuillan, ;Raman spectra of pyridine adsorbed at a silver electrode,; Chemical Physics Letters, vol. 26, pp. 163-166, 1974.
Edgar Guevara, Juan Carlos Torres-Galvan, Miguel G. Ramirez Elias, Claudia Luevano-Contreras, Francisco Javier Gonzalez, “Non-invasive in vivo Raman Spectroscopy of the skin for diabetes screening,” 2017 Photonics North (PN), 2017, pp.1-2, 2017.
A. Mahadevan-Jansen, ;Raman Spectroscopy form Benchtop to Bedside;, in Biomedical Photonics Handbook, T. Vo-Dinh, Ed., London: CRC Press, 2002.
Katerina Hadjigeorgiou, Evdokia Kastanos, Alexandros Kyriakides, Costas Pitris, “Point-of-care Diagnosis of Urinary Tract Infection (UTI) Using Surface Enhanced Raman Spectroscopy (SERS),” 2012 IEEE 12th International Conference on Bioinformatics ; Bioengineering (BIBE), 2012, pp.333-337, 2012.
N. H. Othman, Khuan Y. Lee, A. R. M. Radzol, W. Mansor, N. N. M. Ramlan, “Linear Discriminant Analysis for Detection of Salivary NS1 from SERS Spectra,” TENCON 2017-2017 IEEE Region 10 Conference, 2017, pp.2876-2879, 2017.
Xiaozhou Li, Tianyue Yang, Rong Wang, Weidong Wen, “Surface enhanced Raman spectrum of saliva for detection of lung cancer,” 2011 IEEE International Symposium on IT in Medicine and Education, 2011, vol 2, pp.688-690, 2011.
Yan Wang, Shuang Sun, Dian Qu, Anyu Chen, Zijian Cui, Yulu Yao, Yi Jiao, Xun Guo, Chunwei Liu, “Preliminary study on early detection technology of lung cancer based on surface-enhanced Raman Spectroscopy,” 2010 3rd International Conference on Biomedical Engineering and Informatics, 2015, vol 5, pp.2081-2084, 2010.
P. R. Y. David A. Muller, ;The flavivirus NS1 protein:Molecular and structural biology,immunology, role in pathogenesis and application as a diagnostic biomarker,; 2013.
C. V. Raman, ;A change of wave-length in light scattering 8,; Nature, vol. 121, p. 619, 1928.
Chou, Kuo-Chen; Chen, Nian-Yi (1977). ;The biological functions of low-frequency phonons;. Scientia Sinica, p.447–457.
Chou, K.C. (1989). ;Low-frequency resonance and cooperativity of hemoglobin;. Trends in Biochemical Sciences, p. 212–3.
Gutsche, “Secreted dengue virus nonstructural protein NS1 is an atypical barrel- shaped high-density lipoprotein.” Proc Natl Acad Sci U.S.A, vol 108, no 19, p.8003-8, 2011.
Institute of Photonics, “Surface-Enhanced Raman Spectroscopy (SERS),”2013. Online. Available:http://www.uef.fi/en/web/photonics/sers.
A.James McQuillan, “The discovery of surface-enhanced Raman scattering,” vol63, issue 1, 2009.
Horiba Scientific, “Raman Spectroscopy- SERS substrates,” 2018. Online. Available:http://www.horiba.com/scientific/products/raman-spectroscopy/accessories/sers-substrates/.
A.R.M.Radzol, Khuan Y.Lee, W.Mansor, “Crystalization structure of whole saliva of drop coating deposition Raman for Surface Enhanced Raman Spectroscopy analysis,” 2012 IEEE-EMBS Conference on Biomedical Engineering and Sciences, 2012, pp.438-442, 2012.
Y. Wang, L. Hua, J. Liu, D. Qu, A. Chen, Y. Jiao, X. Guo, C. Liu, W. Huang, and H. Wang, ;Preliminary study on the quick detection of acquired immure deficiency syndrome by saliva analysis using surface enhanced Raman spectroscopic technique,; in Engineering in Medicine and Biology Society, 2009. EMBC 2009. Annual International Conference of the IEEE, 2009, pp. 885-887.
G. Yap, B. K. Sil, and L.-C. Ng, ;Use of Saliva for Early Dengue Diagnosis,; PLoS Negl Trop Dis, vol. 5, p. e1046, 2011.
HYPERLINK ;https://ieeexplore-ieee-org.ezaccess.library.uitm.edu.my/search/searchresult.jsp?searchWithin=;Authors;:.QT.Mohd Azmir Mustapa.QT.;newsearch=true; Mohd Azmir Mustapa, HYPERLINK ;https://ieeexplore-ieee-org.ezaccess.library.uitm.edu.my/search/searchresult.jsp?searchWithin=;Authors;:.QT.Muhammad Hafiz Abu Bakar.QT.;newsearch=true; Muhammad Hafiz Abu Bakar, HYPERLINK ;https://ieeexplore-ieee-org.ezaccess.library.uitm.edu.my/search/searchresult.jsp?searchWithin=;Authors;:.QT.Yasmin Mustapha Kamil.QT.;newsearch=true; Yasmin Mustapha Kamil; HYPERLINK ;https://ieeexplore-ieee-org.ezaccess.library.uitm.edu.my/search/searchresult.jsp?searchWithin=;Authors;:.QT.Amir Syahir.QT.;newsearch=true; Amir Syahir, HYPERLINK ;https://ieeexplore-ieee-org.ezaccess.library.uitm.edu.my/search/searchresult.jsp?searchWithin=;Authors;:.QT.Mohd Adzir Mahdi.QT.;newsearch=true; Mohd Adzir Mahdi, ” HYPERLINK "https://ieeexplore-ieee-org.ezaccess.library.uitm.edu.my/document/8309312/" Bio-Functionalized?Tapered?Multimode?Fiber?Coated?With?Dengue?VirusNS1?Glycoprotein?for?Label?Free?Detection?of Anti-Dengue?Virus?NS1?IgG Antibody,” IEEE Sensors Journal, 2018, vol 18, issue 10, pp.4066-4072, 2018.
W. R. Wong, S. D. Sekaran, F. R. M. Adikan, and P. Berini, “Detection of dengue NS1 antigen using long-range surface plasmon waveguides.,” Biosens. Bioelectron., vol. 78, pp. 132–139, Apr. 2016
P. Wang, G. Brambilla, M. Ding, Y. Semenova, Q. Wu, and G. Farrell, “High-sensitivity, evanescent field refractometric sensor based on a tapered, multimode fiber interference,” Opt. Lett., vol. 36, no. 12, pp. 2233–2235, 2011.
C. V. Raman, ;A change of wave-length in light scattering 8,; Nature, vol. 121, p. 619, 1928.
J. H. Amorim, R. P. Alves, S. B. Boscardin, and L. C. Ferreira, “The dengue virus non-structural 1 protein: risks and benefits,” Virus Res., vol. 181, pp. 53–60, 2014.
R. O. Duda, P. E. Hart, and D. G. Stork, “Pattern Classification,” New York: John Wiley, 2001, pp. 680.
F.M.Twon Tawi, Khuan Y.Lee, W.Mansor, A.R.M Radzol, “Automatic Non structural Protein 1 recignition based on LDA classifier,” 2013 IEEE International Conference on Control System, Computing and Engineering, 2013, pp.340-343, 2013.
Christiani DC, “Genetic susceptibility to lung cancer,” J Clin. Oncol. vol. 24(11), pp. 1651-1652, April 2006.
Fitzmaurice M, Haka A. A, et al. “Raman Spectroscopy: development of clinical applications for breast cancer diagnosis,” SPIE, vol. 1, pp. 5862, 2005.
Tatiana A., Anatoly G. S., Vladimir G. Y., et al. “Positive and negative, drying drops of biological liquids: dynamics of the optical and mechanical properties,” Application in rapid medical diagnostics, SPIE, vol. 188, pp. 5692, 2005.
Neda Baheri, Mohammad Hossein Miran Baygi, Rasoul Malekfar, “Applying Principle Component Analysis for Detecting Skin Damage Caused by Using Detergents: A Raman Spectroscopy Study,” 2009 International Conference on Information Management and Engineering, pp.267-270, 2009.
Weilin Wu, Haiming Gong, Mingyu Liu, Guannan Chen, Rong Chen, “Noninvasive breast tumors detection based on saliva protein surface enhanced Raman spectroscopy and regularized multinomial regression,” 2015 8th International Conference on Biomedical Engineering and Informatics (BMEI), pp.214-218, 2015.
S.Saraswasthi, B.S.Mahanand, A.Kloczkowski, S.Suresh, N.Sundararajan, “Detection of onset od Alzheimer’s disease from MRI images using a GA-ELM-PSO classifier,” 2013 Fourth International Workshop on Computational Intelligence in Medical Imaging (CIMI), pp.42-48, 2013.
S Jerald Jeba Kumar, C G Ravichandran, “Macular Edema severity detection in colour fundus images based on ELM classifier,” 2017 International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud), pp.926-933, 2017.
Vipin K Mishra, Varun Bajaj, Anil Kumar, “Classification of normal, ALS, and myopathy EMG signals using ELM classifier,” 2016 2nd International Conference on Advances in Electrical, Electronics, Information, Communication and Bio-Informatics, pp.455-459, 2016.
Amin Hekmatmanesh, Seyed Mohammad Reza Noori, Mohammad Mikaili, “Sleep spindle detection using modified extreme learning machine generalized radial basis function method,” 2014 22nd Iranian Conference on Electrical Engineering (ICEE), pp.1898-1902, 2014.
K. Sai Deepak and Jayanthi Sivaswamy, Member, IEEE, “Automatic Assessment of Macular Edema From Color Retinal Images,” IEEE Transactions On Medical Imaging, Vol. 31, No. 3, March 2012.
P. Pal, N. Mohanty, A. Kushwaha, B. Singh, et al. “Feature extraction for evaluation of Muscular Atrophy,” IEEE Int. Conf. Computational Intelligence and Computing Research (ICCIC), 2010, pp. 1-4.
A. R. M. Radzol, Khuan Y. Lee, W. Mansor, N. Ariffin, “Biostatistical Analysis of Principle Component of Salivary Raman Spectra for NS1 Infection ,” 2016 IEEE EMBS Conference on Biomedical Engineering and Sciences (IECBES), pp.13-18, 2016.
HYPERLINK ;https://en.wikipedia.org/wiki/Karl_Pearson; o ;Karl Pearson; Pearson, K. (1901). HYPERLINK ;http://stat.smmu.edu.cn/history/pearson1901.pdf; ;On Lines and Planes of Closest Fit to Systems of Points in Space; . Philosophical Magazine, vol 2, no 11, pp. 559–572, 1901.
Shifei Ding, Han Zhao, Yanan Zhang, Xinzheng Xu, Ru Nie, “Extreme learning machine:algorithm, theory and applications,” Artificial Intelligence Review, vol 44, pp.103-115, 2015.