Fu-Malaria

Menu

  • Home
  • About
  • Help Us
  • FU Malaria Background
  • Issues & Limitations

FU Malaria

Open Source Computer Vision Platform for the Detection of Malaria and other Blood diseases.

Star

Fork

Download

Demo

About

FU Malaria - Open Source Computer Vision Platform for the Detection of Malaria and other Blood diseases

Original Code & IP donated by Visimagic.com and Xinova.com

Last year my engineer FV and I (@Mikenicholls88) spent a summer working on a skunk works project to train OpenCV to detect Malaria in blood cells.

Actual working output, blood cells with red border, Malaria shown with Black rings/outlines.

You can see a live demo here FU Working Demo Site

We had a vision of using a low cost computing device like a Raspberry Pi or Intel Joule to power a low cost portable Microscope <$300 that could use computer vision to detect Malaria and other diseases. (Mockup below)

Malaria kills 500,000-750,000 people a year, many of them young children and elderly. It often occurs in remote places that lack access to modern pathology services and so diagnosis is difficult and misdiagnosis is common.

We worked in the startup team of Intellectual Ventures Invention Development Fund and we prototyped a lot of projects but this one wasn't officially sanctioned :) and ended up getting dropped.

The good news is that the new team at http://xinova.com/ have kindly contributed this project to our company Visimagic Inc and we are making it available to the world to help reduce Malaria transmission.

Help Us

We are asking developers in communities around the world to contribute their progamming and computer vision skills to enhance this project.

It should not need to be said but I will say it anyway, we invite anyone from any community, race, sex, sexual orientation or religion to contribute and encourage anyone who wants to try to build a business around this contribution if it helps achieve the aim of reducing Malaria.

FU Malaria Background

FU Malaria is a platform that uses OpenCV and a rule set that detects and mark most Malaria species that affect humans. It displays the image with a colored circles to indicate the blood cells and any Malaria pathogens in an image of a thin blood smear.

Intellectual Ventures has a long track record of helping the Gates Foundation with Malaria projects but this was outside official channels and designed to be a fast and dirty experiment to see if we could train computer vision to detect Malaria in a blood slide image.

We picked this because we wanted to validate if it was possible to put this capability into a small low cost computing package like a Raspberry PI or Intel Joule Computer Vision kit along with a Microscope so that it could go into field hospitals and clinics everywhere in the world to help cut down the transmission vector of the disease. (this is one of the biggest issues, cut the time it takes to diagnose and treat and you cut the multiplier effect of transmission).

After being restructured in 2016 and various life changes my previous employer Invention Development Fund (IDF) one of the investment funds of Intellectual Ventures has very kindly agreed to contribute this code to Visimagic Inc which is taking the step of open sourcing it so that developers of the world who feel strongly about this can do something about it.

Originally implemented on Windows we are moving this to Linux.

Issues & Limitations

This is a good proof of concept. To make successful in the real world it needs to have some rough edges sanded off.

It has the following issues Im aware of

  • It will occassionally detect cell artifacts that are not Malarial.
  • It needs further work on species detection, especially some of the less obvious species.
  • It is currently configured to work on Thin Blood smears, normally a lab will use a thick smear in order to get sufficient volume to ensure sufficient sensitivity, while it does pick up pathogens in Thick smears it is less definitive and further work from Malarial experts is needed to codify the differences between pathogens and artifacts.

Copyright (C) 2015-2016 - Mike Nicholls, FV

Top