Machine-learning technology developed to differentiate types of blood clots based on shape

Platelets (also called thrombocytes) are small sticky cell fragments that circulate through the blood stream and group together to fix damaged blood vessels. This grouping, known as a blood clot, is a biological mechanism to prevent excessive bleeding from injury. However, excessive clotting activity can block blood vessels, causing a heart attack or stroke. There are several causes (agonists) of blood clots, and until now it has been considered impossible to classify the cause of the clots, even through the use of microscopic imaging.

A group of researchers have recently developed a tool that uses machine-learning technology to differentiate different types of blood clots based on subtle differences in shape. Based on this information, physicians can more easily diagnose the cause of individual blood clots as well as how to treat them.

Lead author Yuqi Zhou and her colleagues started by taking blood samples from a healthy individual and exposed them to different clotting agents. These different clotting agents (also called coagulants) are molecules or substances that cause platelets to clot together, such as vasopressin. (On the other end of the spectrum, drugs like aspirin limit platelet signalling so they don’t stick together. People experiencing debilitating blood clots will often also take blood thinners like warfarin and heparin.) The team of researchers captured thousands of images of the different varieties of clots caused by each clotting agent, using a technique called high-throughput imaging flow cytometry.

Next, they used a type of machine-learning technology called a convolutional neural network, or CNN. It starts with an input image, assigns importance (weighting or biasing certain features) to various aspects or objects in the image, and differentiates that image from others. They trained their computer (called the iPAC, intelligent platelet aggregate classifier) to identify subtle differences in the shape of clots caused by different clot agonists. Out of 25,000 clot images that the computer had never seen before, iPAC was able to distinguish most of the clot types. They also tried blood samples from several different people, with the same result.

“Using this new tool may uncover the characteristics of different types of clots that were previously unrecognized by humans, and enable the diagnosis of clots caused by combinations of clotting agents. Information about the causes of clots can help researchers and medical doctors evaluate the effectiveness of anti-clotting drugs and choose the right treatment, or combination of treatments, for a particular patient.”

Keisuke Goda, Senior Author and Professor at the University of Tokyo’s Department of Chemistry

Zhou and her colleagues hope that the newly-developed technology could be used to understand the mechanism behind the strange blood clotting seen in some patients afflicted with COVID-19. However, much about the virus (and its less common symptoms) remains unknown.

References

1. Zhou Y, Yasumoto A, Lei C, Huang C-J, Kobayashi H, Wu Y, Yan S, Sun C-W, Yatomi Y, Goda K. 2020. Intelligent classification of platelet aggregates by agonist type. eLife.
2. Williams M. 2017. What are Platelets and Why are They Important? Johns Hopkins Medicine. Johns Hopkins University.
3. Saha S. 2018. A Comprehensive Guide to Convolutional Neural Networks — the ELI5 way. Towards Data Science. Medium.
4. Blasi T, Hennig H, Summers HD, Theis FJ, Cerveira J, Patterson JO, Davies D, Filby A, Carpenter AE, Rees P. 2016. Label-free cell cycle analysis for high-throughput imaging flow cytometry. Nature Communications 7.