Health-AIM: An artificial intelligence approach for inference with clinical health datasets

In the clinical health domain, artificial intelligence (AI) models need to use data to make reliable decisions regarding patient trajectories and treatments. Incorrect decisions can lead to strain on both caregiver and patient, and clinical datasets often have low volume, high dimensionality, and many missing values. This thesis presents Health-AIM, a Python platform designed to address these challenges by leveraging well-known AI techniques to make effective inferences using clinical health datasets. Health-AIM introduces a four-step workflow consisting of data preprocessing, feature extraction, inferencing, and results visualization, with comprehensive functionalities to handle a variety of scenarios in each step. Health-AIM supports the usage of custom probabilistic graphical models and of classifiers from Python’s sklearn library. Naïve Bayes networks and Bayesian networks specified by edge lists can be dynamically constructed and trained using Health-AIM. Accepting labeled patient data and user-parameters as input, Health-AIM enables runtime selection of data imputation, feature selection, classification, and testing methods with a single function call. Thus, different inferences can be performed rapidly in succession through simple adjustment of parameters. A real-life case study on prediction of postoperative survival in patients with metastatic epidural spinal cord compression is used to illustrate the various functionalities of Health-AIM.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2020

Author dc:creator

• Anjur, Vikram Sriram

Contributors dc:contributor

• Iyer, Ravishankar K
• Arnold, Paul

• data science, artificial intelligence, inference, classification, feature extraction, data imputation, machine learning, statistics, data preprocessing, probabilistic graphical models, clinical health prediction

Statement dc:rights

• Copyright 2020 Vikram Sriram Anjur

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/108322

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/108322