Artificial Intelligence in Hematology
Artificial Intelligence in Hematology: Transforming Diagnostics and Therapeutics.
Introduction:
Artificial intelligence (AI) has emerged as a transformative force in medicine, and hematology is no exception. By leveraging AI, clinicians and researchers can enhance diagnostic accuracy, personalize treatment plans, and streamline workflows. This article explores the current applications, benefits, and future potential of AI in hematology.
Hematology, the branch of medicine concerned with the study, diagnosis, treatment, and prevention of blood diseases, has traditionally relied on microscopic examination and various laboratory tests. However, the integration of AI is revolutionizing these practices, offering unprecedented capabilities in data analysis and pattern recognition.
Peripheral blood smear examination is one of the basic steps in the evaluation of different blood cells. It is a confirmatory step after an automated complete blood count analysis. Manual microscopy is time-consuming and requires professional laboratory expertise. Therefore, the turn-around time for a peripheral smear in a health care centre is approximately 3-4 hours. To avoid the traditional method of manual counting under the microscope a computerized automation of peripheral blood smear examination has been adopted, which is a challenging task in medical diagnostics. In recent times, deep learning techniques have overcome the challenges associated with human microscopic evaluation of peripheral smears and this has led to reduced cost and precise diagnosis. However, their application can be significantly improved by the availability of annotated datasets.
Example of annotations in a color image (top row) and a monochromatic stack (bottom row). The three channels that constitute the stack are shown separately as images (A) (transmission), (B) (DAPI), and (C) (red fluorescence channel). Only the cells that are present in the field of view are labelled, according to the legend’s pseudo-colors.
AI algorithms can analyze vast amounts of data, including patient records, laboratory test results, and imaging scans, to identify patterns and anomalies that may be indicative of hematological conditions. This data-driven approach can lead to more accurate and earlier diagnoses, ultimately improving patient outcomes.
Over the last 15 years, comprehensive diagnostics in leukemia and lymphoma has become increasingly challenging. In order to follow the guidelines of the World Health Organization (WHO) classification, the results from different fields, including cytomorphology, cytogenetics, immunophenotyping, and molecular genetics, have to be combined to establish a diagnosis. Gains in throughput from the introduction of next-generation sequencing (NGS) technologies and the accompanied broadening of the analytical spectrum in molecular genetics have boosted the value of molecular genetic results for diagnostics, as indicated by the revision of the WHO classification of leukemias and lymphomas in 2017.
The plethora of available molecular information has broadened the landscape in leukemia and lymphoma diagnostics and has led to new insights in the underlying biology of the respective diseases, provoking a shift in diagnostics from phenotype to genotype. Moreover, the identification of an increasing list of diagnostic and prognostic markers, the refined estimate of inter-individual variability, and the ongoing effort to establish correlations between different layers of information that might eventually lead to improved targeted therapy options, are paving the way for personalized medicine.
In parallel, it is indisputable that the data collection process goes digital, allowing the automated integration of different test results and easy access for all involved stakeholders. This journey also offers the opportunity to share information between clinical and genomic experts from multiple institutions, facilitating the assignment of patients to specific clinical trials or targeted treatment options. Hence, the journey goes from analogous to digital and from phenotype to genotype.
Digital data is also a basic prerequisite for the application of emerging artificial intelligence (AI) techniques. Together with deep learning (DL) and machine learning (ML), AI is currently a buzzword across almost all scientific disciplines and has the potential to revolutionize diagnostic approaches in hematology. With the dramatic performance improvements in the last years, AI is at the brink to be introduced into routine diagnostics to enhance diagnostic methods but even more to facilitate disease classification and guidance of treatment.
Due to the widespread interest and success of AI-based applications, the terms: artificial intelligence and machine learning, resound throughout the varying scientific disciplines, while often being used interchangeably in medicine. However, whereas AI strives to simulate human behavior and intelligence, ML, as a subdomain of AI, refers to the automatic detection of patterns and associations within the data. DL, as a subfield of ML, allows layered neural networks to learn an abstract representation of often very complex data sets.
The left side represents the different domains of supervised learning going from artificial intelligence to machine learning and finally deep learning. The right side depicts the process of supervised learning. At the top right corner, the requirements for a good training data set are listed. The data is used for automated feature extraction, leading to the generation of a model, the performance of which is evaluated by its capability to correctly predict the labels of unseen instances (= test data). Based on the evaluation outcome, the model is retrained to refine the selected features and to optimize the model. After several rounds of retraining the final model emerges. AI artificial intelligence, DL deep learning, ML machine learning.
One exciting prospect is the development of digital twins to forecast cancer trajectories and to predict the potential impact of different therapeutic strategies in silico. The evaluation of these simulations might help to select the most promising interventions for each individual patient, minimizing side effects and the risk of complications.
AI in Hematologic Diagnostics:
Automated Blood Smear Analysis
One of the primary applications of AI in hematology is the automated analysis of blood smears. Traditional manual examination of blood smears is time-consuming and subject to human error. AI algorithms, particularly deep learning ones, can analyze blood smear images with high precision. These systems can identify and classify different types of blood cells, detect abnormalities, and even recognize rare diseases.
Among the most promising applications of AI in hematology is the automated analysis of blood smears and other hematological images. AI-powered image recognition algorithms can accurately identify and classify different types of blood cells, including red blood cells, white blood cells, and platelets. This technology can streamline the process of cell counting and morphological analysis, reducing the workload of hematologists and providing more consistent and reliable results.
For instance, convolutional neural networks (CNNs) have shown remarkable accuracy in detecting conditions such as leukemia and malaria from blood smear images. Studies have demonstrated that AI can match or even surpass the diagnostic capabilities of experienced hematologists in certain scenarios.
Two primary cell detection and classification types of algorithms were developed. One was trained with color images, and the other one was trained with the three-channel monochromatic stack (transmission, DAPI, red, TDR). The annotations made on the color images were copied to the TDR image set. Consequently, both algorithms contain the same number of images and annotations.
Example of the automatic detection of blood cells in color images (first and third row) and TDR stacks (second and fourth row). TDR images are transmission, DAPI and red-fluorescence channels merged into one. The rainbow-colored areas with thin white borders represent objects (cells) that had been automatically identified by the algorithm. Red areas in rows two and four without white borders represent the red fluorescence of the Giemsa stain.
Flow Cytometry Data Interpretation
Flow cytometry is a key clinical tool in the diagnosis of many hematologic malignancies and traditionally requires close inspection of digital data by hematopathologists with expert domain knowledge. Advances in artificial intelligence (AI) are transferable to flow cytometry and have the potential to improve the efficiency and prioritization of cases, reduce errors, and highlight fundamental, previously unrecognized associations with underlying biological processes.
Over the last decade, artificial intelligence (AI) has advanced significantly. In one study , the researchers developed and validated an AI-assisted flow cytometry workflow using 379 clinical cases from 2021, employing a 3-tube, 10-color flow panel with 21 antibodies for primary immunodeficiency diseases and related immunological disorders. The AI software (DeepFlow™, version 2.1.1) is fully automated, reducing analysis time to under 5 min per case. It interacts with hematopathologists for manual gating adjustments when necessary. Using a proprietary multidimensional density–phenotype coupling algorithm, the AI model accurately classifies and enumerates T, B, and NK cells, along with important immune cell subsets, including CD4+ helper T cells, CD8+ cytotoxic T cells, CD3+/CD4−/CD8− double-negative T cells, and class-switched or non-switched B cells. Compared to manual analysis with hematopathologist-determined lymphocyte subset percentages as the gold standard, the AI model exhibited a strong correlation (r > 0.9) across lymphocyte subsets. The study highlights the accuracy and efficiency of AI-assisted flow cytometry in diagnosing immunological disorders in a clinical setting, providing a transformative approach within a concise timeframe.
Comparison of MFC plots generated by KaluzaTM and AI DeepFlowTM for normal T, B, and NK cells.
To summarize, flow cytometry is a crucial technique in hematology for characterizing cell populations. AI tools can process the complex data generated by flow cytometry, identify cell populations more accurately, and predict disease subtypes. Machine learning algorithms, including random forests and support vector machines, are being employed to enhance the analysis of flow cytometry data, leading to quicker and more accurate diagnoses.
AI in Hematologic Therapeutics
Predictive Analytics for Treatment Response:
AI-based predictive models can analyze a patient’s genetic and clinical data to determine the risk of developing hematological disorders, such as leukemia or anemia. By identifying high-risk individuals, healthcare providers can implement preventive measures and personalized treatment plans, tailored to the patient’s unique characteristics and disease profile. This personalized approach has the potential to improve treatment outcomes and reduce the burden of hematological diseases.
AI’s predictive capabilities are particularly beneficial in personalized medicine. By analyzing large datasets of patient information, AI can predict how patients will respond to specific treatments. For example, AI models can forecast responses to chemotherapy in leukemia patients, enabling personalized treatment plans that optimize efficacy while minimizing adverse effects.
Drug Discovery and Development:
AI accelerates the drug discovery process by identifying potential therapeutic targets and predicting the effectiveness of new compounds. In hematology, AI has been instrumental in discovering novel treatments for conditions such as multiple myeloma and lymphoma. AI-driven simulations and models help understand these diseases’ molecular mechanisms and design targeted therapies.
Improved Clinical Decision-Making:
Artificial intelligence can also assist hematologists in making more informed clinical decisions. AI algorithms can integrate and analyze a wide range of data, including laboratory test results, patient histories, and treatment outcomes, to provide healthcare providers with evidence-based recommendations for diagnostic tests, therapeutic interventions, and disease management strategies.
Enhancing Hematology Workflows
Laboratory Automation:
AI-driven automation is transforming hematology laboratories. From sample processing to result interpretation, AI systems enhance efficiency, reduce turnaround times, and minimize human error. Automated systems for tasks such as complete blood count (CBC) analysis are becoming increasingly prevalent, allowing hematologists to focus on more complex diagnostic and therapeutic decisions.
Telemedicine and Remote Diagnostics:
AI-powered telemedicine platforms are expanding access to hematologic care, particularly in remote and underserved regions. These platforms use AI to analyze patient data and provide diagnostic insights, enabling remote consultations and reducing the need for in-person visits. This is especially beneficial for chronic hematologic conditions that require ongoing monitoring.
Opportunities and Outlook:
As indicated before, the best opportunities for a fast implementation of ML-based technologies into routine diagnostic workflows offer cytomorphology and cytogenetics. With the recent integration of DL methods, the accuracies of the different methods are close to expert level, offering the possibility of faster and more accurate sample processing, bringing expertise to the fingertips of less experienced hematologists. Although the current AI-based applications in flow cytometry are less extensive and have not reached clinically acceptable accuracies in all domains, an automated workflow could potentially lead to more standardized and reproducible results.
Overview of implemented and potential ML applications in hematology. The central part of the figure displays current and future diagnostic tests and methods, while the outer part illustrates the various data types and the potential clinical impact of ML-based applications and analyses. DL deep learning, ML machine learning.
Challenges and Ethical Considerations:
While the potential of AI in hematology is immense, several challenges and ethical considerations must be addressed. Data privacy and security are paramount, given the sensitive nature of medical information. Moreover, the integration of AI into clinical practice requires rigorous validation and regulatory approval to ensure patient safety and efficacy. Ethical concerns also arise regarding the transparency of AI algorithms and the potential for bias in AI-driven decisions.
Conclusion:
AI is poised to revolutionize hematology by enhancing diagnostic accuracy, personalizing treatments, and improving workflow efficiency. As AI technologies continue to evolve, their integration into hematologic practice will undoubtedly lead to better patient outcomes and more efficient healthcare delivery. However, careful consideration of ethical and regulatory challenges is essential to realise AI’s benefits in haematology fully.
References:
Walter, W., Haferlach, C., Nadarajah, N. et al. How artificial intelligence might disrupt diagnostics in hematology in the near future. Oncogene 40, 4271–4280 (2021). https://doi.org/10.1038/s41388-021-01861-y. https://www.nature.com/articles/s41388-021-01861-y
Swerdlow SH, Campo E, Harris NL, Jaffe ES, Pileri SA, Stein H et al. editors. WHO classification of tumours of haematopoietic and lymphoid tissues. Revised 4th edn. Lyon: International Agency for Research on Cancer; 2017.
D. Connie Garzon – What is Artificial Intelligence (AI) and Why People Should Learn About it. https://incubator.ucf.edu/what-is-artificial-intelligence-ai-and-why-people-should-learn-about-it/
Madhavan S, Beckman RA, McCoy MD, Pishvaian MJ, Brody JR, Macklin P. Envisioning the future of precision oncology trials. Nat Cancer. 2021;2:9–11.
Björnsson B, Borrebaeck C, Elander N, Gasslander T, Gawel DR, Gustafsson M, et al. Digital twins to personalize medicine. Genome Med. 2019;12:4.
Fjelland R. Why general artificial intelligence will not be realized. Humanit Soc Sci Commun. 2020;7:10.
Campos-Medina, M.; Blumer, A.; Kraus-Füreder, P.; Mayrhofer-Reinhartshuber, M.; Kainz, P.; Schmid, J.A. AI-Enhanced Blood Cell Recognition and Analysis: Advancing Traditional Microscopy with the Web-Based Platform IKOSA. J. Mol. Pathol. 2024, 5, 28-44. https://doi.org/10.3390/jmp5010003
Shams UA, Javed I, Fizan M, Shah AR, Mustafa G, Zubair M, Massoud Y, Mehmood MQ, Naveed MA. Bio-net dataset: AI-based diagnostic solutions using peripheral blood smear images. Blood Cells Mol Dis. 2024 Mar;105:102823. doi: 10.1016/j.bcmd.2024.102823. Epub 2024 Jan 4. PMID: 38241949.
Campos-Medina, M.; Blumer, A.; Kraus-Füreder, P.; Mayrhofer-Reinhartshuber, M.; Kainz, P.; Schmid, J.A. AI-Enhanced Blood Cell Recognition and Analysis: Advancing Traditional Microscopy with the Web-Based Platform IKOSA. J. Mol. Pathol. 2024, 5, 28-44. https://doi.org/10.3390/jmp5010003
David P. Ng, Paul D. Simonson, Attila Tarnok, Fabienne Lucas, Wolfgang Kern, Nina Rolf, Goce Bogdanoski, Cherie Green, Ryan R. Brinkman, Kamila Czechowska
Recommendations for using artificial intelligence in clinical flow cytometry – Ng – Cytometry Part B: Clinical Cytometry – Wiley Online Library https://onlinelibrary.wiley.com/doi/10.1002/cyto.b.22166
Lu, Z.; Morita, M.; Yeager, T.S.; Lyu, Y.; Wang, S.Y.; Wang, Z.; Fan, G. Validation of Artificial Intelligence (AI)-Assisted Flow Cytometry Analysis for Immunological Disorders. Diagnostics 2024, 14, 420. https://doi.org/10.3390/diagnostics14040420
Smith, J., et al. (2022). “Deep Learning for Automated Blood Smear Analysis.” Journal of Hematology and Oncology, 15(3), 345-357.
Johnson, L., et al. (2021). “AI in Hematologic Diagnosis: Current Applications and Future Directions.” Blood Advances, 5(4), 123-134.
Kumar, R., et al. (2020). “Machine Learning in Flow Cytometry: Enhancing Hematologic Diagnostics.” Clinical Cytometry, 72(6), 567-578.
Davis, T., et al. (2019). “Predictive Analytics in Leukemia Treatment: An AI Approach.” Hematology Reports, 11(2), 98-107.
Garcia, M., et al. (2023). “AI in Drug Discovery for Hematologic Malignancies.” Future Oncology, 19(1), 12-25.
Lee, S., et al. (2021). “Automation in Hematology Laboratories: AI-Driven Advances.” Laboratory Medicine, 52(4), 456-467.
Patel, N., et al. (2020). “Telemedicine and AI in Hematology: Bridging the Gap.” Telemedicine Journal and e-Health, 26(7), 789-798.
Rodriguez, H., et al. (2019). “Ethical Considerations in AI-Driven Hematology.” Journal of Medical Ethics, 45(3), 203-210.
Saumyab. Machine Learning vs Deep Learning vs Artificial Intelligence. https://www.analyticsvidhya.com/blog/2021/06/machine-learning-vs-artificial-intelligence-vs-deep-learning/
Gavita Regunath. Understanding The Difference Between AI, ML, And DL: Using An Incredibly Simple Example. https://www.advancinganalytics.co.uk/blog/2021/12/15/understanding-the-difference-between-ai-ml-and-dl-using-an-incredibly-simple-example
Keywords:
Artificial Intelligence, Hematology, Blood Smear Analysis, Flow Cytometry, Predictive Analytics, Drug Discovery, Laboratory Automation, Telemedicine, Ethics in AI, Image Analysis, Predictive Modeling, Personalized Treatment, Clinical Decision-Making.
