In 2013, a computer systems engineer by the name of Hisashi Kambe was working on an innovative system. It was called BakeryScan and used a neural network model called AlexNet, which employed a technique called deep learning to classify images into different object categories.
It was a good fit. Japanese bakeries pride themselves on a wide variety of pastries. Bread and pastries have long been an import product in Japan, with scarcity and lack of variety being commonplace for centuries. As a result, people cherish variety now, and studies suggest that in Japan, the more types of pastries you have, the more you sell. In fact, many bakeries in Japan have a constant influx of new pastries coming in and out of selection. So much so, in fact, that even employees have a hard time knowing what some items are.
Having an AI that could automatically identify pastries makes things much easier. It enables fast, accurate recognition of pastries at checkout, reducing errors and speeding up the customer service process. It can also help with inventory management by tracking sales and product availability, ensuring that popular items are restocked efficiently. The system’s adaptability allows it to handle diverse product shapes and sizes, making it a valuable tool in busy bakery environments.
It worked well. It became pretty popular in the pastry industry, selling well, but then things took an unexpected turn.
From bread to healthcare
In 2017, a doctor in Kyoto, saw an ad for the algorithm. He thought the pastries looked pretty similar to cancer cells.
Thus began BakeryScan’s foray into the cancer detection world. In 2018, at a conference in Sapporo about the AI identification of cancer cells, Kambe said the algorithm is still impractical for some tasks due to the large amounts of data required. But he continued working on it — and deep learning algorithms also improved significantly.
By 2021, a prototype of BakeryScan customized for cancer detection (called Cyto-AiSCAN) was being trialed in two major hospitals in Japan. The algorithm had vastly improved — it didn’t need to look at single cells, it could work with entire microscope slides. Kambe refused to elaborate on how exactly the algorithm works, saying it’s an “original way,” just like with the pastries.
But the results spoke for themselves: the algorithm reportedly had 98% accuracy.
What started as an innovative solution for a bakery problem had evolved into a breakthrough in medical diagnostics. Hisashi Kambe’s journey with BakeryScan is a story of scientific success and adaptability, showing how a system designed for something as simple as identifying pastries could be repurposed for life-saving technology.
Nowadays, there are several AIs capable of medical diagnoses, but this is more than just a tale of deep learning or AI’s potential. It’s a testament to the flexibility of science. Since then, BakeryScan has been adapted to distinguish between pills in hospitals, amulets sold by shrines, and even 18th century Japanese woodblock prints.
