Machine learning identifies bugs that spread Chagas disease


New research from the University of Kansas shows that machine learning is able to identify insects that spread the incurable disease called Chagas with high accuracy based on common digital photos. The idea is to give public health authorities where Chagas is a new tool to contain the spread of the disease and eventually offer identification services directly to the general public.

Chagas is particularly unpleasant because most people who have it do not know they have been infected. But according to the Centers for Disease Control and Prevention, about 20-30 percent of the 8 million chagas worldwide are hit at some point by heart rhythm abnormalities that can cause sudden death; dilated hearts that do not pump blood efficiently; or a dilated esophagus or colon.

The disease is most often caused when the triatomines – more commonly known as "kissing insects" – sting people and transmit the parasite Trypanosoma cruzi to the bloodstream. Chagas is most prevalent in rural areas of Mexico, Central and South America.

A recent venture in the KU, called the Virtual Vector Project, sought to allow public health officials to identify triatomines carrying Chagas with their smartphones, using a kind of portable photo studio to take pictures of the insects.

Now a KU graduate student has built this project with a proof-of-concept research showing that artificial intelligence can recognize 12 Mexican and 39 Brazilian species of kisses with high precision by analyzing common pictures – an advantage for authorities who want cutting spread of Chagas' disease.

Ali Khalighifar, a KU doctoral student at the Biodiversity Institute and the Department of Ecology and Evolutionary Biology, led a team that has just published Journal of Medical Entomology. To identify insects who kiss regular shots, Khalighfar and his colleagues worked with Google's deep open source learning software called TensorFlow, which is similar to the technology behind Google's reverse image search.

Because this model is able to understand, based on tones and pixel colors, what is involved in an image, you can take the information and analyze it in a way the model can understand – and then you provide it for testing and testing can identify them with a really good identification rate. That's without preprocessing – you just start with raw images, which is incredible. That was the goal. Previously, it was impossible to do the same thing accurately and certainly not without preprocessing the images. "

Ali Khalighifar, PhD student at the KU Institute of Biodiversity and the Department of Ecology and Evolutionary Biology

Khalighifar and his co-authors РEd Komp, researcher at the KU Information and Telecommunications Technology Center, Janine M. Ramsey of the National Institute of Public Health of Mexico, Rodrigo Gurgel-Gon̤alves of the University of Brasilia, and A. Townsend Peterson, KU Distinguished Professor of Ecology and Evolutionary Biology and senior curator of the KU Biodiversity Institute Рtrained their algorithm with 405 images of Mexican triatomines and 1,584 images of Brazilian triatomines.

At first, the team was able to achieve "83.0 and 86.7% of correct identification rates in all Mexican and Brazilian species, respectively, an improvement over comparable rates for statistical classifiers," they write. But after adding information on the geographical distributions of insect kisses in the algorithm, the researchers increased the accuracy of identification to 95.8% for Mexican species and 98.9% for Brazilian species.

According to Khalighifar, algorithm-based technology could allow public health authorities and others to identify triatomine species with an unprecedented level of accuracy to better understand disease vectors in the soil.

"In the future we hope to develop an application or a web platform of this model that is constantly trained based on the new images, so it is always being updated, which provides high quality IDs for any interested user in real time," he said.

Khalighifar is now applying a similar approach using the TensorFlow for instant identification of mosquitoes based on the sounds of their wings and frogs based on their calls.

"I'm working on mosquito recordings now," he said. "I have switched from recordings to signal images, so we used TensorFlow to identify the other one." "I've changed the processing of images to signal the processing of the recordings of mosquito wings. project I am working on now are frogs with Dr. Rafe Brown of the Biodiversity Institute, and we are designing the same system to identify these species based on the calls given by each species.

Although artificial intelligence is often portrayed popularly as a threat to kill jobs or even an existential threat to humanity, Khalighifar said his research showed how AI could be of benefit to scientists studying biodiversity.

"It's incredible – AI really is capable of doing everything, for better or for worse," he said. "There are uses that appear to be frightening, such as identifying Muslim faces on the streets." Imagine if we can identify frogs or mosquitoes, how easy it is to identify human voices. trying to use the good side of this technology to promote biodiversity conservation and support public health work. "


Journal Reference:

Khalighifar, A. et al. (2019) Deep Learning Algorithms Improve Automated Identification of Chagas Disease Vectors. Journal of Medical Entomology.


Source link