Machine Learning solution to improve care management for pneumonia

The Project

Currently, diagnosing pneumonia is slow, and it involves disconnected systems, and many people including radiologists, doctors, and technicians. Inefficient processes and disconnected systems cause issues including provider stress, low productivity, inaccurate results, and delayed care, which leads to poor patient outcomes. To solve this, I have developed an automated machine-learning solution on AWS that can detect pneumonia quickly and accurately. This makes results more reliable and also reduces stress on doctors. These advantages will result in fewer deaths, lower cost, and leads to a faster cure. These issues affect providers and patients! Not only does the patient needing help currently get affected, but the patients waiting in line for help also have to wait longer! My project is to create a machine learning solution on AWS (Amazon Web Services) to detect and improve care for pneumonia patients by delivering results faster and more accurately. This solution is developed utilizing AWS Cloud with services to store, query, orchestrate, process, and notify. Healthcare companies can ingest DICOM (Digital Imaging and Communications in Medicine) images into AWS storage and that will trigger workflows to detect and notify regarding patient conditions using Machine Learning. This solution provides better and faster care for pneumonia. FULL RESEARCH PAPER: https://wp.me/peEpEY-8

View this project on YouTube

Team Comments

Every year more than 2 million pneumonia patients are dying worldwide, and in fact, my little sister also has pneumonia. So we need an intelligent system to detect pneumonia very quickly and accurately. For these reasons, I decided to build a solution for pneumonia to help patients worldwide.

One of the hardest parts of this project were integrating different services together. It took much troubleshooting, learning, and looking at examples along the way to complete the project. Once I figured out how the services worked together, they seemed to get along much more effortlessly.

My next steps are to upgrade the ML model and the solution to work with more than one disease. I would also try to make this solution faster and reduce and eliminate inefficiencies. In addition to that, I would create my infrastructure via Infrastructure as Code, and improve data security.