Analysis of inflammation, metabolic and clinical markers in predicting fat mass in HIV-positive males using artificial neural network / Nurul Farhah Shamsuddin

Nurul Farhah, Shamsuddin (2021) Analysis of inflammation, metabolic and clinical markers in predicting fat mass in HIV-positive males using artificial neural network / Nurul Farhah Shamsuddin. Masters thesis, Universiti Malaya.

[img] PDF (The Candidate’s Agreement)
Restricted to Repository staff only

Download (718Kb)
    PDF (Thesis M.A)
    Download (1217Kb) | Preview


      Increased inflammation has been discovered in people living with Human Immunodeficiency Virus (HIV), which has been associated with the development and advancement of metabolic diseases. Immune activation and inflammation have been revealed to affect the redistribution of fat in people living with HIV (PLWH) whether the individual is on treatment with antiretroviral therapy (ART) or treatment naïve. Therefore, the assessment of body fat composition of PLWH based on markers of inflammation is important to develop keen diagnosis and prognosis of medical conditions facing by these patients, for example lipodystrophy syndrome, metabolic syndrome, dyslipidemia and diabetes mellitus. The purpose of this study is to investigate the relationship among inflammation marker, interleukin-6 (IL-6), clinical and metabolic factors with fat mass (FM) in males living with HIV (MLWH) using artificial neural network (ANN) and to assess the contribution of metabolic and clinical factors alongside inflammation marker, IL-6 to body fat composition in MLWH. Five sets of ANN models were constructed, and each set of network model was produced and manipulated based on inclusion of inflammation marker, IL-6 as an independent variable and on statistically significant association of independent variables with dependent variable, FM. First set of ANN models used inflammation, metabolic and clinical variables as input, the second set of ANN models used metabolic and clinical variables as input, the third set of ANN models used body mass index (BMI) and waist-hip-ratio (WHR) as input, the fourth and fifth set of ANN models used statistically significant parameters associated with FM but each set applied different training split ratio. Dependent variable consisted of body composition variable, FM collected from whole-body DEXA scan was selected as output. Comparison of model performance were assessed through the model error function, error sum of squares (SSE), relative error (RE) and mean predictive accuracy percentage (MPA%). The MPA% obtained by Set A was 84.84% with ten hidden nodes in its single-hidden layer ANN, 84.02% with ten hidden nodes for Set B, 80.09% with eight hidden nodes for Set C, 85.26% with four hidden nodes for Set D and 85.48% with two hidden nodes for Set E. Through independent sample t-test conducted on Set A and Set B ANN performance to discover the influence of the inflammation marker, IL-6, no significant difference was found in the MPA% between ANN models using clinical, metabolic and inflammation data and ANN models using clinical and metabolic variables, t (13) = 0.75, p< .05, 95% C.I. [-0.95% – 1.96%]. However, statistically significant difference in prediction accuracy was found on fat mass in MLWH at the p <.05 level for the five ANN models sets [F (4, 45) = 8.802, p = 0.00]. The findings show that ANN technique was able to triangulate the relationship between body fat composition and clinical, metabolic and inflammation data of MLWH even though the addition of inflammation marker, IL-6 did not significantly improve the ANN performance.

      Item Type: Thesis (Masters)
      Additional Information: Dissertation (M.A.) - Faculty of Engineering, Universiti Malaya, 2021.
      Uncontrolled Keywords: Artificial neural network (ANN); Body composition; HIV; Inflammation; Metabolism
      Subjects: T Technology > TJ Mechanical engineering and machinery
      Divisions: Faculty of Engineering
      Depositing User: Mrs Rafidah Abu Othman
      Date Deposited: 29 Apr 2022 01:09
      Last Modified: 29 Apr 2022 01:09

      Actions (For repository staff only : Login required)

      View Item