AI Analysis of Red Blood Cells: A Q&A with Dr. Ahmed Elnady Elsafty

The development of artificial intelligence (AI) presents exciting opportunities in the pathology and laboratory landscape. Automating the analysis of red blood cell (RBC) morphology is one area where AI can greatly benefit laboratory professionals in a time of critical employee shortages.   

Critical Values spoke with Ahmed Elnady Elsafty, MD, CLS(ASCPi)SH, SC, PGDHHM, CEO of PathOlOgics, LLC. Dr. Elsafty is the lead author of a study published in Nature where he and his colleagues created an extensive dataset of RBC morphologies for deep learning automation. His goal is to develop AI that can mimic experts’ observations of blood smears, saving valuable time.  

Here, Dr. Elsafty shares his experience creating a dataset of one million segmented RBCs, and his goals for AI collaboration.  

What was the impetus for your research? 

Right now, we don’t have a way to automate RBC morphology analysis, even though the need is vital given a growing shortage of pathologists. Teardrop-shaped RBCs and fragmented RBCs are often associated with medical disorders that could be fatal. And increased ovalocytes (oval-shaped RBCs) are associated with almost all types of anemia.  

Unfortunately, no existing art or solution on the market meets hematopathologists’ needs for clinically usable, automated analysis of manually prepared and stained blood smears. 

In layman’s terms, what were your findings? 

With an accuracy level exceeding 99% and minimal false positives/negatives, we’ve provided 1,000,000 segmented RBCs, including 240,000 classified into nine shapes and 47,000 patches from 25 manual blood smears. This extensive dataset supports the development of an app that would process field images from light microscopes or digital pathology scanners, delivering percentages of clinically significant RBC shapes in just seconds. Such an app could mimic expert observation; as soon as the stain and smear are suitable for the human eye, the app could also analyze it. 

How will this research impact pathology and laboratory medicine? 

There’s a significant need for AI assistance given the anticipated critical shortage of about 5,000 to 5,700 pathologists in the United States by 2030, coupled with a continued need for 500 million complete blood count (CBC) tests annually in the U.S. and over 3.1 billion globally.  

Our goal is to assist hematologists in overcoming stress and fatigue, saving time, addressing the shortage of experts, and increasing productivity with minimal liability. This application would work with commonly used, manually prepared and stained blood smears without requiring any prior standardization of staining or smearing procedures. 

What is your next step? Are you seeking collaboration on your dataset?  

Yes, our immediate next step is to evaluate the app across multiple medical centers to assess how its performance compares to experts. Once it’s confirmed reliable for RBC morphology analysis, we can proceed with additional studies to determine cutoffs and perform analyses for different RBC types, particularly ovalocytes, in various grades of anemia (low RBC count) and erythrocytosis (high RBC count.) This could lead to a sensitive screening test for anemia and erythrocytosis across all severity grades and types.  

Additionally, it could provide a specialized diagnostic tool for detecting and counting true teardrop-shaped RBCs and schistocytes (fragmented) RBCs, valuable in both blood banks and clinical care settings, especially before platelet transfusions. If the provided dataset proves insufficient, we are committed to supplying an additional dataset, published publicly, which will include any missed, mis-segmented, or misclassified cells. This will be based on evaluations from hundreds or thousands of manually prepared blood smears, aiming to achieve a verified world-class standard of accuracy and reliability. 

What do you hope this leads to?

We have both our technological and medical wings working hard to advance our technology to the level of senior hematopathologists, so we can offer valuable assistance with white blood cells (WBCs) too. For WBC analysis, we are collecting 10,000 bone marrow samples from all hematologic disorders, including comprehensive confirmatory tests such as flow cytometry. We are seeking grants and financial support to make the WBC analysis publicly available, similar to our achievement with RBCs. 

Is there anything else you'd like to add?  

Soon we will publish a multicentric study evaluating the first phase of our web-based automated narrative clinicopathologic correlation builder for hematopathology. This tool integrates qualitative and quantitative results from CBC, smear reviews, flow cytometry, cytogenetics, and molecular testing to generate a narrative clinicopathologic correlation that mirrors the final reports of senior hematopathologists, verified against WHO 2022 standards. Please visit https://cbctst.com/Analysis/Create to learn more.