This project was worked on as part of CPSC 490: Special Projects, the senior special project requirement for Computer Science majors. I worked with Allen Wang ’18 and Irene Li (PhD candidate), under the supervision of Professor Dragomir Radev. Abstract, full report, and code repo presented below:

Abstract

This paper explores approaches to designing and implementing a computer system that allows patients with cough to diagnose themselves without input from a medical expert. After performing literature review and studying tangential research efforts, we explore approaches to modelling the diagnosis process. This involves patient, symptom, and diagnosis representation, as well as building a prototype that encodes and utilizes relevant expertise to provide reasonable diagnoses to the user. To tackle the complex task of diagnosing an arbitrary patient, we simplify the problem to patients with a given “principal symptom”, specifically cough.

We first attempt to build a simple classification model using machine learning techniques. However, this proves to be an extremely challenging task, given the lack of sufficient, real-life, well-formatted data, and the absence of a reliable mechanism for us to collect such data. Even learning paradigms that require minimal training data fail to produce any meaningful results, given the large number of parameters of even the simplified version of the problem. As a result, we transition to a probabilistic model, and encode domain knowledge in the form of medical expertise (approximating statistical evidence).

Our prototype overcomes the challenge of lacking realistic data by relying on statistical evidence that is widely collected and easily retrievable. In addition, it greatly facilitates extensibility and refinement of the model, to allow for further experimentation and potential future use. Though this work focuses on patients with cough, the approaches discussed, software implementation, and conclusions drawn can be generalized to apply to other principal symptoms, and thus to generic patients.

Project Report: Final Project Report

Github link: A System to Diagnose Patients with Cough