Diagnosing pneumonia in under-resourced countries

Contact: mheal-pneumonia@umich.edu


Abstract

Among all infectious diseases, pneumonia is the single largest source of mortality in children under the age of five.  Pneumonia is also particularly fatal for infants; 80% of these deaths come from children under the age of two.  A more effective method for pneumonia detection in these countries would improve accuracy of diagnosis, which would reduce the pneumonia mortality rate in these regions. We want to overcome this by creating a device to accurately diagnose pneumonia in low-resource regions. Many researchers have been taking steps towards developing methods of computationally detecting pneumonia. They have explored machine learning, signal processing and statistical methods as diagnostic tools.  Our team plans on building off of their previous work in order to create an accurate and robust device. Our system will use machine-learning based classification models of waveform data (e.g., respiratory sounds, thoracic vibrations, sound permeation, etc.) in order to distinguish between patients with and without pneumonia. Many researchers have explored this method, and our research has concluded that this will be the most effective way to create a low cost and accurate diagnostic tool. Our team has decided that a device that uses signal processing to analyze breathing sounds, then classify them using a machine learning base classification system is the most effective way to diagnose childhood pneumonia.


Problem

Pneumonia is the single largest source of mortality in children, as it takes the lives of over one million children each year. To further complicate the situation, unrefined diagnostic techniques lead to cases of pneumonia being diagnosed as malaria in young children in under-developed countries. 

OUR PROJECT

Team Pneumonia aims to create an aggregated diagnostic tool or device which will aid those in under-developed countries to attain an accurate measurement and diagnosis of pneumonia, preventing a misdiagnosis as malaria or other bronchial diseases. Our tool will be intuitive to use, allowing for a preliminary diagnosis outside of a strict medical setting which can be difficult to come by for many who are susceptible to pneumonia. Our team is currently in the process of researching the most accurate diagnostic tests for pneumonia identification and we hope to develop models for a prospective device soon. 

 
 

Project Leads

Zane Dunnings | zldunn@umich.edu