DeepHeme, an artificial intelligence tool, helps to automate the counting of different blood cell types contained in a blood or bone marrow sample.
Researchers at Memorial Sloan Kettering Cancer Center (MSK), the University of California, San Francisco (UCSF), and University of California, Berkeley (UC Berkeley) have developed an artificial intelligence (AI) tool called DeepHeme that can help automate the diagnosis of blood and bone marrow cancers. The study was published on June 11 in Science Translational Medicine.
Traditionally, diagnosing these cancers has required doctors to manually count and classify hundreds of cells under a microscope — a labor-intensive process used to detect disease and determine cancer stage. DeepHeme uses AI to perform this task with expert-level accuracy, reducing the time required from more than 30 minutes to just seconds. The model was trained on nearly 50,000 annotated digital cell images. It was then tested on unseen cases and found to match or exceed the performance of expert pathologists.
Dr. Gregory Goldgof is is Director of AI and Computational Hematopathology at MSK.
“DeepHeme can analyze both blood and bone marrow samples and may support future efforts to improve personalized medicine,” says MSK computer scientist and pathologist Gregory Goldgof, MD, PhD, MS, senior author of the study. “Automation with DeepHeme will support large-scale research efforts to develop tools that better predict which patients will respond best to different treatment options.”
MSK plans to begin using DeepHeme in the clinic after further validation and may license it to other hospitals. This work is part of MSK’s broader digital pathology and AI initiative, which aims to accelerate and enhance cancer diagnosis using cutting-edge computational tools.
AI Improves Many Areas of Cancer Care
Researchers and technologists across MSK are looking for ways to harness the power of AI to improve patient care. For example, artificial intelligence is already useful for interpreting mammograms, analyzing large amounts of genomic data, and reviewing patients’ self-reported side effects from treatment.
AI tools are not meant to replace human expertise but instead to improve the speed and accuracy of clinical analysis and help identify patterns across large numbers of patients and diverse disease types.
How Digital Pathology Can Improve Patient Care
DeepHeme is part of a greater effort to make MSK the first fully digital hematopathology service in the country — relying primarily on digitized slides instead of traditional glass slides and microscopes. Hematopathologists are doctors who study slides and images of blood and bone marrow samples.
Dr. Ahmet Dogan is Chief of MSK’s Hematopathology Service.
Physician-scientist Ahmet Dogan, MD, PhD, Chief of MSK’s Hematopathology Service, explains that having digital versions of slides provides many other benefits beyond their use in AI. For example, the slides don’t need to be physically transported to a pathologist’s location — they can be reviewed on a computer screen anywhere. Images can also be included in a patient’s medical record.
“If a patient has follow-up testing, even years after an earlier biopsy, we’ll be able to instantly pull up all their slides and compare them,” Dr. Dogan says. “We don’t have to take the old slides out of storage.”
Dr. Goldgof, who is Director of AI and Computational Hematopathology at MSK, adds that DeepHeme also may help identify new biomarkers that are based on morphology, or the way cells look under the microscope. Biomarkers are measurable characteristics of cells that could help to detect and classify disease, determine how a patient is responding to treatment, or predict how likely it is that a patient’s disease will progress. Finding biomarkers is important for advancing the field of personalized medicine.
“There are 500 million blood samples collected in the United States every year,” Dr. Goldgof says. “We could never analyze all of them with humans, but we could do it with AI. This could lead to many important future applications.”
Digital Efforts Drive Research Across MSK
DeepHeme is only one of many exciting new research efforts at MSK to deploy the latest digital tools. In addition to guiding patient care, digital versions of pathology images give scientists the data they need for large research programs like MSK’s Cancer Data Science Initiative, part of the Marie-Josée and Henry R. Kravis Center for Molecular Oncology. This effort collects millions of anonymized data points from patients across MSK, in near real time as they are being treated.
The massive database created by this initiative is a goldmine for researchers seeking to understand which treatments work best for which patients and under what circumstances. While the data cannot be traced back to an individual patient, the insights can help ensure that current and future patients get the right treatment for their individual cancer.
The Department of Pathology and Laboratory Medicine is a driving force behind these efforts. “DeepHeme is just one example of MSK’s broader digital transformation,” Dr. Dogan says. “Now that we have high-resolution scans of every slide, we can use them to learn more about blood cancers in general.”
It’s important to note that AI will never be a replacement for the analysis done by doctors, he explains: “The diagnosis of cancer is a medical decision that needs to be made by a human. You need doctors and other researchers to continually improve the algorithms and adapt them to new clinical questions.”
