Reading Exercise | | The Pastry AI that Learned to Fight Cancer

Japan’s long history of trade is considered one of the reasons behind the country’s very diverse food tastes. Because of this, unlike French or Italian bakeries that offer only a few options, Japanese bakeries offer pastries of all sizes, shapes, flavors, and colors. They have options like The Carbonara, which is a pastry version of the famous Italian pasta dish, or The Ham Corn, a breakfast pastry topped with ham, corn, and mayonnaise. There are hundreds of different types of pastries in these unique bakeries. Unfortunately, this diversity did not come without a cost: cashiers had to spend months learning the price of each individual pastry based on sight alone. This meant that the checkout process was not only very difficult for cashiers, but also caused long wait times for customers.

A software company called Brain was asked to help resolve the problem of confusion at the cash register. Brain, which was founded by computer programmer and software designer Hisashi Kambe, had always worked on projects based on computer visualization capabilities. The company originally designed computers that could detect errors in formulas for fabric patterns, so resolving the problem of visualizing hundreds of different pastries was no stranger to them. Brain began working on a software called BakeryScan.

BakeryScan is unique because, unlike deep learning software like Google Translate, Siri, and almost every AI system out there, BakeryScan doesn’t need large amounts of specialized data to make decisions; it is created to understand irregularities like the shadow cast into the middle of a donut hole or the slightly darker color of over-baked bread without needing the input of tens of thousands of similar images.

Once BakeryScan was implemented, it became a hit. It was televised all over Japan and became such a cultural phenomenon that it was even referenced in their language proficiency exams.

This was how a doctor at the Louis Pasteur Center for Medical Research, in Kyoto, saw a television segment about the machine. He realized that cancer cells, under a microscope, looked a lot like bread. He contacted Hisashi Kambe’s company Brain to see how they could collaborate to develop a version of the program that could help pathologists detect cancer cells. BakeryScan was already equipped with tools that allowed human experts to give the program feedback, the only thing they needed to change was what exactly the system would be analyzing.

They started small, analyzing single cells under a microscope, but eventually moved up to more complex images. Now, BakeryScan, adapted and renamed Cyto-Aiscan, is being tested in two major hospitals in Kobe and Kyoto. It is capable of “whole-slide” analysis, meaning that more than analyze a single cell at a time, it is capable of looking at an entire microscope slide and identifying the cells that might be cancerous. Instead of considering the shadows cast into a donut hole or the darker shade of over-baked bread, the software is now considering the color tone of the nucleus, its size and texture, and its overall roundness.

Who knew that the world of pastries could bring us further ahead in cancer research?

Did you enjoy reading this article? Test your reading skills by completing the quizzes below.