We use automated image analysis powered by machine learning to facilitate diagnosis of parasitic worm infection in the developing world. We designed our tool to be coupled with inexpensive and highly mobile cell phone microscopes that can be operated by almost anyone. ParasiteID reduces the bottleneck to patient treatment by removing the need for highly trained microscopists and the expensive microscopes conventionally used in diagnosis.
We have created a highly sensitive and specific diagnostic tool by assembling the largest helminth egg image dataset of its kind and by implementing cutting edge deep learning (machine learning) techniques.
This project began as a capstone project for the UC Berkeley Master of Information and Data Science program.
The ParasiteID Team thanks Dr. Danielle Skinner and Dr. Conor Caffrey (UC San Diego) for generously preparing and sharing Schistosome eggs. We also thank Dr. Margaret Mentink-Kane (NIH Schistosome Research Center) for providing Schistosome eggs. We thank Dr. John Quinn (Makerere University, Uganda), Dr. Kamarul Hawari Ghazali (Universiti Malaysia Pahang), and Alicia Alva and Dr. Mirko Zimic (Universidad Peruana Cayetano Heredia, Lima, Peru) for sharing images.
Open-source negative images were taken from ImageNet (http://www.image-net.org/), and we made heavy use of the 2014 and 2016 winning models, VGG16 and resnet50, respectively.
The ParasiteID Team also thanks Dr. Mike Tamir, Dr. Jimi Shanahan, Oscar Holmstrom, Jackson Bell, and Dr. Isaac Bogoch for advice regarding ParasiteID development. Finally, we thank Joyce J. Shen and David Steier for their invaluable guidance and support.
We stand on the shoulders of giants! Code was inspired by, adapted from, and used from the following sources:
We are also appreciative of the great institutions who have made their material free and open-source: