When did it start and how is it evolving?
Artificial Intelligence (AI) seems to be a new tool in the world, but if we look into its development, we can be surprised to know that AI was created around 1950. It was inspired by the goal of developing machine-learning systems with architectures that resemble aspects of the human brain, an approach that later contributed to the development of deep learning and, eventually, large language models (LLMs). In the early 2000s, thanks to the generation of more potent computers, an increase of AI tools was developed to generate specialized tools for specific tasks. For example, AI tools would be trained to analyze human tissues, including Mammography Intelligent Assessment (MIA) to process mammographies, or ENDO-AID CADe/CADx for polyp detection in endoscopy videos. With the evolution and improvement of these AI tools, their aims also evolved, and some generative AI (e.g. ChatGPT) have allowed a broader and more accessible use of AI.
If we observe AI tools improvement and how they became generalized instead of only specialized, we notice that the learning method is different. The early learning method would happen through personalized information and training acquired from the same people who developed the AI tools. This has changed recently as AI tools have gained context through a multitude of available information such as databases, websites, and books.
AI applications in the medical area
AI use in the medical area has exploded in recent years, either for image analysis from several tissues, for aiding in diagnostic or for assisting in administrative tasks.
There have been published studies that demonstrate the efficiency and accuracy of these foundation models in different medical fields. For example, Kermany, DS et al. reported that the model they developed was able to analyze 100.000 retina pictures with supervised learning, and the model could have an extremely high accuracy compared to humans1. However, how the models acquired a better accuracy, and how it reached that stage, is still not well defined. Furthermore, it has been reported that retinal pictures can lead to the discovery of a wide set of diseases such as Alzheimer's2, Parkinson's3, and diabetes4. Interestingly, this is not only true for retinal pictures. It was shown that cardiograms can be used to identify a wide variety of diseases not directly related to the heart, including diabetes, which before the use of AI tools was not possible.
Current applications of AI in hematology
In hematology, AI is primarily used as a tool to classify and analyze patients' data. It has been extensively applied to process cytogenetic, cytomorphology (Cellvision) and flow cytometry data obtained from blood and bone marrow samples. Regarding flow cytometry, a study by Mocking R. T., et al. demonstrated that AI tools can assess measurable residual disease (MRD) in patients with acute myeloid leukemia, and that the use of this tool was not only faster than humans, but also improved the accuracy of MRD assessment 5.
A new avenue has been pursued by using AI to classify cancer patients based on their transcriptomics. Beder T. et al. have trained a machine-learning classifier, named ALLCatchR, to use RNA-seq gene expression data with the aim of identifying molecular subtypes and classifying them into WHO/ICC subtypes6. Overall, different types of data from patients have been used to train machine learning models to improve AI tools’ performance and to improve patient diagnosis.
What are the advantages?
While we inform ourselves about what machine learning systems or AI tools can accomplish, we are almost marveled by their accuracy and their rapidity in data analysis. Several studies in the medical field, including in hematology, have reported that AI tools can reduce the time of analysis from minutes to a few seconds compared to humans, while providing higher rates of accuracy. In addition, by using AI tools, clinicians would be able to reduce the time spent on administrative tasks and the writing of medical reports. Taking all this into account, it suggests that AI tools can improve diagnostics and patient treatments, time management of clinical staff, and ultimately patients' outcomes.
What are the limitations and problems?
One of the major limitations of using AI tools is the amount of patients' information that is shared and the struggle of how to handle it while remaining ethical towards patients' privacy. Another major limitation is the data quality used to train AI models, as their performance and reliability are as good as the quality of the data it's given for training.
Additionally, there is not enough information about how AI tools generate their outcome from the medical analysis, or questions we might ask. We train the system and we obtain a response, but still, the process by which this response was generated remains elusive most of the time. In relation to this, different outcomes have been shown between human analysis and AI tools. For example, even for a seemingly straightforward task such as RT-qPCR primer design, human expertise remains more reliable than AI-generated solutions.
Furthermore, several researchers and clinicians share the concern that the critical thinking of our and future generations might be reduced because of the extensive use of AI tools to help with sample analysis, diagnostics, or even treatments. Lastly, the problem of accountability has emerged among clinicians and researchers. If the diagnosis is incorrect or the treatment did not work as a result of the AI response, who is accountable for this, and how will we evaluate it?
Future of AI
The way AI has evolved strongly suggests that it has become, and will remain, an integral part of research and data analysis. By enabling faster and more accurate processing of large and complex datasets, AI has the potential to significantly enhance biomedical research. At the same time, biomedical professionals remain cautious, implementing safeguards to better understand AI decision-making processes, validate its outputs, and protect patients' private data. Currently, researchers and clinicians are working closely with AI developers to address critical unanswered questions, particularly those that require the analysis of vast amounts of data.
While significant challenges remain, the future of AI in hematology, but also in medicine overall, appears promising, provided that its adoption is guided by transparency and reliability.
Figure created with ChatGPT
References:
1. Kermany DS, Goldbaum M, Cai W, Valentim CCS, Liang H, Baxter SL, McKeown A, Yang G, Wu X, Yan F, Dong J, Prasadha MK, Pei J, Ting MYL, Zhu J, Li C, Hewett S, Dong J, Ziyar I, Shi A, Zhang R, Zheng L, Hou R, Shi W, Fu X, Duan Y, Huu VAN, Wen C, Zhang ED, Zhang CL, Li O, Wang X, Singer MA, Sun X, Xu J, Tafreshi A, Lewis MA, Xia H, Zhang K. Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning. Cell. 2018 Feb 22;172(5):1122-1131.e9.
2. Cheung C, Ran A, Wang S et al. A deep learning model for detection of Alzheimer's disease based on retinal photographs: a retrospective, multicentre case-control study. The Lancet Digital Health, 2022; 4, e806-e815
3. Li SWR, Gardner A, Votruba M. The Retina as a Biomarker for Parkinson's Disease: A Systematic Review. Neuroophthalmology. 2025 Aug 14;50(1):1-12.
4. Poplin, R., Varadarajan, A.V., Blumer, K. et al. Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nat Biomed Eng 2, 158–164 (2018).
5. Mocking, T.R., Haaksma, L.H., Reuvekamp, T. et al. Computational measurable residual disease assessment in acute myeloid leukemia: a retrospective validation in the HOVON-SAKK-132 trial. Leukemia 39, 2559–2562 (2025).
6. Beder T, Hansen BT, Hartmann AM, Zimmermann J, Amelunxen E, Wolgast N, Walter W, Zaliova M, Antić Ž, Chouvarine P, Bartsch L, Barz MJ, Bultmann M, Horns J, Bendig S, Kässens J, Kaleta C, Cario G, Schrappe M, Neumann M, Gökbuget N, Bergmann AK, Trka J, Haferlach C, Brüggemann M, Baldus CD, Bastian L. The Gene Expression Classifier ALLCatchR Identifies B-cell Precursor ALL Subtypes and Underlying Developmental Trajectories Across Age. Hemasphere. 2023 Aug 25;7(9):e939.
Blog post contributed by Alexandra Bacquelaine Veloso and Camille Malouf of the ISEH New Investigators Committee.
Please note that the statements made by Simply Blood authors are their own views and not necessarily the views of ISEH. ISEH disclaims any or all liability arising from any author's statements or materials.