A good deal of clinical diagnostics are effectively performed by cytologists who examine cells through a microscope for signs of disease. This is an imperfect, slow process that depends on the training, focus, and attention to detail of the cytologist. Now, researchers at Helmholtz Zentrum München and the University Hospital of LMU Munich in Germany have developed an automated system that points to the reality of cytologists becoming an endangered species.
They taught a computer, running deep learning algorithms, to automatically classify cells within blood samples for signatures of acute myeloid leukemia (AML). This required a collection of nearly 20,000 images of individual blood cells, some of which were obtained from patients with AML. The computer was eventually able to define the variables that point to diseased cells and the researchers confirmed the technology by running images of cell samples from 100 AML patients and 100 healthy control subjects through the detection system. The same images were inspected by a group of professional cytologists.
The results showed that the computer was at least as capable as trained humans at classifying AML cells. Since the technology consists of a software algorithm, it should be easy to roll it out in hospitals around the world. Moreover, it points to the ability of the same approach to diagnose many other diseases that cytologists are typically involved in detecting.
“To bring our approach to clinics, digitization of patients’ blood samples has to become routine,” said Dr. Carsten Marr, the lead researcher, in a press release. “Algorithms have to be trained with samples from different sources to cope with the inherent heterogeneity in sample preparation and staining. Together with our partners we could prove that deep learning algorithms show a similar performance as human cytologists. In a next step, we will evaluate how well other disease characteristics, such as genetic mutations or translocations, can be predicted with this new AI-driven method.”
Image: The deep learning algorithm classifies leukocytes in a blood smear in an automated and standardized way. Left: What human experts classify. Right: Pixels important for AI analysis. © Helmholtz Zentrum München / Carsten Marr
Study in Nature Machine Intelligence: Human-level recognition of blast cells in acute myeloid leukaemia with convolutional neural networks
