Non-invasive dengue screening method via optical spectroscopy: A multivariate investigation / Abdul Halim Poh Yuen Wu

• Main Author: Abdul Halim Poh , Yuen Wu
• Format: Thesis
• Published: 2019
• Subjects: R Medicine (General); TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://studentsrepo.um.edu.my/13135/1/Abdul_Halim_Poh.pdf; http://studentsrepo.um.edu.my/13135/2/Abdul_Halim_Poh.pdf; http://studentsrepo.um.edu.my/13135/

More feared than understood, dengue fever often evokes emotions of morbidity in the public space, arguably due to a positive response to media campaigns. A symphony of clinical analysis of blood results (not necessarily dependent only on the NS1 detection), symptoms, and day of fever onset, among others, contributes to a successful diagnosis, often spanning many man-hours (or days) combined with costly overhead expenditure. For dengue-related management, the Government of Malaysia spent close to US$73.5 million in 2010, which was 0.03% of the country’s GDP during that time. With the aim of reduction of the time and monetary costs involved in this issue, a potentially non-invasive method was scrutinized in painstaking detail to circumvent at least a few hurdles in dengue patient care, especially in diagnostics. This method is namely Diffuse Optical Reflectance Skin Spectroscopy. With a probe and a light source combined with spectrometers, the forearm of a suspect patient in the University of Malaya Medical Centre (UMMC) is scanned. The spectroscopy data is then collected and saved via a software tailored to consolidate all information regarding the patient. The patient is then assigned to a diagnosis by UMMC physicians. Three groups were later formulated for the classification, namely confirmed dengue, probable dengue, and control patients. Based on multivariate analysis on the spectroscopy data of 230 patients, we have come to at least two major findings. First, the modelling produced by the statistical algorithms predicted the accuracy of detection up to 98.65% on discriminating all three groups. Second, several feasible algorithmic models for classifying dengue patients was synthesized, ranging from sensitivities and specificities of 76.67%-89.29% and 94.85- 100% respectively. From these findings, further clinical trials on non-invasive dengue screening are recommended.