How Did A Bakery Lead To Advanced Medical Imaging Equipment?

Technology evolves in rather unique ways and for rather unusual reasons, and nothing embodies this principle in medical imaging more than how an early AI-driven cancer detector originated as a way to tell pastries apart.

To explain why, it is important to understand the need for an AI scanner for bakeries in the first place.

Starting in the mid-2000s, artisan bakeries and patisseries were becoming very popular in Japan but quickly became a victim of their own success when people queued up around the block to pick up cakes, sandwiches and pastries, leaving the staff overwhelmed.

Automation could help here, but fresh pastries could not have a barcode applied to them, and wrapping them up could make it appear as though they are not actually fresh baked goods.

The solution, therefore, was BakeryScan, an AI scanner that can detect the type and quantity of baked goods on a tray, identifying often variable products by common key characteristics.

Once sufficiently trained, BakeryScan was remarkably accurate, even compared to other machine learning algorithms, and this led a doctor working for the Louis Pasteur Center for Medical Research in Kyoto to have a moment of inspiration.

Cancer cells viewed on a slide via a biopsy do not look entirely dissimilar to baked goods, so if the underlying technology could detect different types of bread, perhaps it could identify different types of cancer.

This led to the development of AI-Scan, a more variable form of the technology that could learn very quickly the different types of cells that could be cancerous from training data and provide at least a starting point if not an outright confirmation for a doctor by scanning an entire slide at once.

The system did not work the same way as modern deep learning and machine learning algorithms, but it highlighted that the biggest benefit of AI-powered medical imaging equipment was not just accuracy but speed.