File(s) under embargo

11

month(s)

6

day(s)

until file(s) become available

INTELLIGENT HEALTHCARE DATA ANALYTICS COUPLED WITH SENSOR ASSESSMENT FOR NON-ALCOHOLIC FATTY LIVER DISEASE (NAFLD)

thesis
posted on 03.05.2022, 13:20 by Ridhi DeoRidhi Deo

This research was conducted to develop and evaluate a screening tool for Hepatic Steatosis (or fatty liver) detection using machine learning based models. The developed models are intended to be used as a potential clinical decision support tool for identifying patients with Non-Alcoholic Fatty  Liver  Disease  (NAFLD).  Two  versions  of  a  HS  prediction  tool  are  discussed  in  Paper  1, Objectives 1A, and 1B, respectively.

Explainability  analysis  of  the  developed  models  is  also  a  major  component  of  this  work, discussed in Paper 2. Models from Paper 1 are analyzed further for interpretability and the results are then compared with current clinical literature. Insights from the explainability analysis are used to  identify  best  models  that  follow  the  clinical  literature  logically.  Most  contributing  features within each model are also identified in this work.

Another aspect of NAFLD management is related to the chronic exposure to heavy metals in the environment (such as: Arsenic, Lead, Cadmium etc.). The heavy metal exposure component is explored in two ways in this dissertation. In paper 3, another version of the ML-based screening tool is explored by including heavy metal exposure data. The results from the model (with heavy metal data) are then compared with models that exclude the heavy metal exposure data. The results and their implications are discussed in paper 3.

Arsenic is a major hepatotoxin and the chronic exposure can lead to severe liver injury. In Paper  4,  a  commercially  available  Arsenic  detection  kit  was examined  for  Arsenic  detection  in water  at  a  household  level.  The  kit  was  evaluated  following  a  short  experimental  plan  and  the obtained  results  are  discussed.  Finally,  the  obtained  images  were  quantified  digitally  using  a customized image analysis and pattern recognition algorithm. The methods used for quantification and the obtained results are also discussed.


Funding

Ross Fellowship

History

Degree Type

Doctor of Philosophy

Department

Technology

Campus location

West Lafayette

Advisor/Supervisor/Committee Chair

Suranjan Panigrahi

Additional Committee Member 2

Edward A. Liechty

Additional Committee Member 3

Jennifer L. Freeman

Additional Committee Member 4

Frederick C. Berry