Home / Blog / AI bests the best human biochemist

AI bests the best human biochemist

Human induced pluripotent stem cell neurons imaged in phase contrast (gray pixels, left). The image is currently processed manually with fluorescent labels (color pixels) to make them visible. Credit: Google

Every day petabytes of medical images are created around the world. Biochemists, radiologists and doctors are looking at these images to detect what “might” be wrong.

Researchers are looking at this trying to find ways for automating the process. Image recognition by computers has been the holy grail of artificial intelligence and in these last decade has seen tremendous advances.

A joint cooperation of Harvard, Gladstone Institute and Google has led to the creation of a system for detecting cells, their type and structure in terabytes of bioimages created every day at the Gladstone Institute using deep learning. This technology can recognise patterns and make prediction.

Tests have shown the system has an accuracy of up to 98%, a human biochemist expert has 80%. AI is beating us in yet another area.

The learning process used images of cells as they were created by the microscope and then the same images labelled by a biochemist using fluorescent dies. Millions of these images have been used to train the system and then the systems started to operate on its own on the images produced by the microscope.

As learning engine the researchers used TensorFlow, an open software learning tool developed by Google. This is very interesting because it shows not just the power and capability of the system, it also shows that learning engines can be created by adapting existing ones and applied to a variety of areas. Since we are now harvesting more and more data, continuously, in different fields, from production to retail, from education to entertainment we have the possibility of leveraging on these data to have a system learning and most importantly predicting what might be next with very high accuracy.  The nice thing is that the accuracy will keep growing since the system will keep learning from new data and from its mistakes.

According to the researchers the whole area of biochemistry can be revolutionised by deep learning making it possible to test drugs with higher accuracy since it will be clear the types of target cells and hence the effects can be associated to those specific cells. What we call today a “disease” is most of the times a cluster of diseases having similar symptoms but actually rooted in different causes, usually difficult to sort out. With deep learning it gets possible to pinpoint the exact cause (genome related) and therefore the association between a drug and a cause is made clear.

About Roberto Saracco

Roberto Saracco fell in love with technology and its implications long time ago. His background is in math and computer science. Until April 2017 he led the EIT Digital Italian Node and then was head of the Industrial Doctoral School of EIT Digital up to September 2018. Previously, up to December 2011 he was the Director of the Telecom Italia Future Centre in Venice, looking at the interplay of technology evolution, economics and society. At the turn of the century he led a World Bank-Infodev project to stimulate entrepreneurship in Latin America. He is a senior member of IEEE where he leads the New Initiative Committee and co-chairs the Digital Reality Initiative. He is a member of the IEEE in 2050 Ad Hoc Committee. He teaches a Master course on Technology Forecasting and Market impact at the University of Trento. He has published over 100 papers in journals and magazines and 14 books.