The nice people at the American Society of
Hematology sent another update today. I don’t post about many of their updates
here as most of them aren’t really aimed at me, but one little snippet of this
update made me think…
Much of what I do each day involves sitting at
a microscope passing judgement on what I think I see down it. I think it fair
to say I have a fair idea what I’m looking at. I think it fair to say that most
people peering down them do. However there is still an awful lot of
subjectivity involved. I can remember my first day on the microscope bench when
one inexperienced trainee MLSO (as we were back then) found something she
couldn’t identify in a blood film. What was that cell she saw? There were as
many opinions as there were people to offer them.
And that is just in peripheral blood. Bone
marrow is far more involved and examination of smears of that stuff requires serious
experience.
Over the years flow cytometry has taken some of
the guesswork out of the matter, and it would seem that neural networks are
possibly the way forward
- A data set of >170 000 microscopic images
allows training neural networks for identification of BM cells with high
accuracy.
- Neural networks outperform a feature-based
approach to BM cell classification and can be analyzed with explainability and
feature embedding methods.
I bet they ain’t cheap… However I will comment
that this sort of thing has been talked about for years and still hasn’t
appeared in routine haematology labs
Highly accurate differentiation of bone marrow cell morphologies using deep neural networks on a large image data set
| Artificial
intelligence (AI) and machine learning algorithms are changing many
facets of our lives and have great potential for disease diagnosis. In
this Plenary Paper, Matek et al describe training a convolutional neural
network model using 171 374 bone marrow cytology images from 945
patients with various hematological diseases before validating its
accuracy in classifying single cells in an independent set of 627
images. The system can automatically classify 22 classes of bone marrow
leukocytes, and with further improvements, it may bring AI-aided
diagnosis of bone marrow biopsy specimens closer to fruition. |
|
No comments:
Post a Comment