For years, I remembered it this way: “The following year, a university and a diagnostics manufacturer published the same idea.” Then old blog entries and public records revealed a slightly different timeline.
01 / WHAT IT IS
First: what is
white-blood-cell morphology?
Blood contains red cells, white cells, platelets, and other components. A white-blood-cell differential classifies the white cells by type, measures their proportions, and checks for abnormal morphology.
Instruments can classify many cells automatically. When abnormal cells are suspected, however, experienced medical laboratory scientists may still need to examine microscope images. It is highly specialized work in which expertise directly supports the result.
02 / WHY AI
The goal was not to use AI.
It was to deliver results sooner.
We did not begin simply because AI was fashionable. We wanted to deliver accurate laboratory information as quickly as possible to physicians treating people with blood disorders. Image classification looked like a way to support the judgment of experienced specialists.
An IBM representative understood the value of creating a medical-AI use case in western Japan and worked internally to secure support for the validation. The platform we chose was then called IBM PowerAI Vision.
We did not want new technology for its own sake.
We wanted to return time to the people waiting for results.
03 / PROOF OF CONCEPT
November 2019.
We showed white blood cells to AI.
We tested whether AI could classify morphology from images of white blood cells. A PoC—proof of concept—is a focused experiment conducted before full product development to answer one question: can this idea work?
The PoC was completed in November 2019. In the assessment of our internal laboratory specialists, it demonstrated classification capability comparable to that of an experienced technologist. The experts who saw the result were genuinely surprised.
cell classification?
Apply image recognition to a task that depends on expert knowledge
practical potential
Not a claim of clinical deployment, but a strong reason to continue validation
04 / SHELVED
The future was visible.
But it never reached production.
The technical potential was clear. A change in organizational direction, however, meant there would be no next validation phase and no production rollout. After roughly six months of work, the initiative stopped there.
What remained was the frustration of hearing that it was not usable. There was no paper, no product, and no public presentation. As the years passed, the work began to feel as though it had never existed outside our memories.
05 / EXCAVATION
The search engine corrected
the person who lived it.
In 2021 I learned about research by Juntendo University and Sysmex and felt that we had missed our chance to be first. I even wrote in my blog that “the same work became a paper the following year.”
The official record tells a different story. Their joint research began in 2016, and a paper on automating blood-cell morphology analysis was published on September 16, 2019—about two months before our PoC was completed. My memory that we had missed a Japanese first was not accurate.
We were not the first. To be honest, that reduced the regret a little. It did not reduce the achievement. Far from a major university or diagnostics manufacturer, we identified the same direction from day-to-day laboratory work, brought IBM and clinical specialists together, and made the idea work as a PoC.
What the excavation uncovered was not an imaginary “first.” It was something more reliable: the ability to begin with a real problem, find the right technology, connect the right people, and turn an idea into something that runs. That is the work I have done again and again.
FACT SOURCES Juntendo University | Topics / Juntendo University | Publications / Juntendo University | 2021 release
We were not beaten to the future.
We had found the same future from the field.