By Renaldo Pool, BHSc
Current clinical laboratory practices, specifically for complete blood count (CBC) analysis, focus on implementing the gold standard of morphological review and interpretation as an initial approach for determining a potential diagnosis (Matek et al., 2021). Despite the increase in multidisciplinary approaches integrated with optical microscopy, the latter is still the preferred method for identifying hematological abnormalities by following the WHO guidelines (Haider et al., 2022; Walter et al., 2023).
However, conventional methods are considered labor-intensive and time-consuming. In addition, the interpretation of cellular morphology and white blood cell differential counts is operator-dependent and influenced by the healthcare professional’s experience and knowledge (Deshpande et al., 2021).
The risk of misidentifying abnormal blood cells and reporting incorrect results significantly affects the turnaround time from specimen collection to result verification (Mutema et al., 2021). Failure to detect abnormalities during manual blood smear reviews can jeopardize patient outcomes. Additionally, when automated analyzers flag a patient's CBC data, daily manual peripheral smear reviews are often performed due to established protocols. These procedures can be time-consuming and may be considered unnecessary in some instances (Zhu et al., 2022).
This article explores how AI/ML tools can be leveraged and integrated into the clinical laboratory, reducing the need for manual peripheral smear reviews and streamlining repetitive tasks. By introducing AI/ML to the healthcare sector, diagnostic accuracy and precision can be achieved through advanced pattern recognition and predictive modeling, contributing to healthcare professionals' rapid intervention and overall patient care. Allow yourself to be immersed in the future of diagnostic hematology!
You will come across the following key concepts in the article where AI/ML enhances the efficiency of CBC analysis:
AI/ML algorithms possess predictive capabilities and utilize pattern recognition through convolutional neural networks (CNNs) to identify complex cellular patterns in blood morphology.
Diverse and inclusive datasets are relied on for practical AI/ML model training to identify different hematological conditions accurately.
With AI/ML implementation, processes are streamlined, and the workflow is optimized with automated detection of hematological conditions.
It enables the automation of routine tasks, decreasing the need for consistent human intervention and decreasing manual interpretive CBC results and human error factors.
AI/ML models are created to continuously learn from new dataset inputs, contributing to a clinical laboratory system that remains updated and capable of evolving with ongoing advancements in hematological diagnostics.
Pattern recognition of blood components
Existing CBC analyzers use measurement principles to provide automated detection of red blood cells, white blood cells, and platelets based on cell size and complexity, i.e., granularity and nucleic complexity inside the cell. The analyzers flag potential abnormalities that necessitate a manual peripheral blood examination. This warrants a review of the peripheral smear by the healthcare professional to confirm whether something is amiss.
AI/ML can recognize complex patterns in large datasets through deep learning models such as convolutional neural networks (CNNs). A large, high-quality training dataset is essential for successfully integrating CNNs into CBC analysis (Matek et al., 2021; El Alaoui et al., 2022). These intricate networks can identify the slightest alterations in cell size, shape, and color, indicating underlying health concerns that traditional analysis tools may overlook.
With AI/ML intervention, different blood components can be recognized accurately to identify normal and abnormal blood cells. Another helpful aspect is examining morphological features, particularly in complex samples with atypical or diverse cell populations. This simplifies the identification of specific indicators of hematological conditions within the CBC results, aiding diagnostic accuracy and precision (El Alaoui et al., 2022).
Moreover, these tools can apply the same criteria across analyses of different patients' CBC results, ensuring uniformity and reliability for each analysis (Meiseles et al., 2022).
Predictive capabilities of AI/ML
Access to a patient’s historical CBC data and clinical history allows AI/ML to undergo predictive modeling training. This technique forecasts a patient’s condition by identifying risk factors associated with specific blood parameter results. These predictive tools enable early intervention and provide a possible prognosis for the patient’s clinical outcome. They also offer the potential for personalized treatment options (El Alaoui et al., 2022; Yang et al., 2024).
A study by Haider et al. (2022) revealed that incorporating AI/ML in hematology diagnostics allows for trend analysis, establishing a ‘disease signature’ or ‘fingerprint’ based on morphological features. It further provides the necessary tools for making predictions and an early diagnostic approach to hematologic conditions.
Furthermore, real-time processing of CBC specimens through AI/ML models provides prompt insights and supports accurate clinical decision-making. This is especially advantageous in the emergency care setting, ensuring the availability of accurate and precise results while aiding in rapid diagnostic applicability.
Workflow optimization in the clinical laboratory
The clinical laboratory workflow is optimized with the addition of rapid anomaly detection using AI/ML algorithms, ensuring that abnormal cells and results are identified accurately and appropriately flagged. The patient’s measured CBC, historical, and clinical data are further enhanced by trend analysis, indicating potential health threats or condition development (Herman et al., 2021). AI/ML methods, such as Large Language Models (LLMs), also improve this by predicting patient deterioration and worsening prognoses. By analyzing historical CBC data combined with clinical information, a forecast of the patient’s health trajectory is created by building a virtual version or digital twin of the patient (Makarov et al., 2024). It provides a proactive approach in real-time, allowing for the rapid detection of cases requiring immediate intervention. Furthermore, severe cases are prioritized for prompt investigation, instigating the need for a personalized treatment plan for the patient (Herman et al., 2021).
This aids healthcare professionals in focusing more effectively on managing patients with critical conditions. AI/ML's predictive functions are interconnected with its ability to constantly learn from new data inputs as new information or cases are presented to the algorithms. This system's adaptive qualities will allow it to evolve continuously with changes in clinical practice and patient outcomes, ensuring CBC analysis's reliability, precision, and accuracy over time (Alanzi et al., 2023).
AI/ML and CBC instruments can also be integrated with existing laboratory information systems (LIS), elevating the data flow and reporting processes. Alongside LIS integration, the availability of patient electronic health records (EHR) further optimizes the clinical workflow, making patient data easily accessible to healthcare professionals (El Alaoui et al., 2022). The AI/ML-generated predictions for a patient’s clinical conditions and outcomes are readily accessible to healthcare professionals, ensuring accurate and efficient decision-making for effective clinical outcomes (Herman et al., 2021).
Decreasing human intervention
Implementing AI/ML tools reduces time-consuming, routine tasks and the need for manual interpretation of CBC results. It also reduces the human error factor when manual methods are required. This is because specific manual tasks are time-consuming and operator-dependent; that is, each individual’s skill set and experience vary (Herman et al., 2021; Obstfeld, 2023). Reducing the requirement for consistent human intervention, especially for mundane tasks, allows healthcare professionals to prioritize process improvement, research, and development.
The technical foundations of AI/ML
Due to variations occurring in CBC data, AI/ML implements different strategies to ensure uniformity and standardization.
Pre-processing data
Normalization and standardization techniques are used to address variations seen in data. Data is scaled to a similar range and centered to generate a mean of zero and a standard deviation of one. If outliers are present, they are identified and managed through these algorithms. Outliers are characteristically known as data points that significantly differ from others or an entire data set. AI/ML models can remove or adjust outliers' influence on the general analysis by applying a Z-score or interquartile range. This ensures that data is not biased or skewed, providing a well-represented population (El Alaoui et al., 2022; Katare et al., 2022; Rodgers et al., 2023).
Combined algorithm use
Using robust algorithms has a profound impact because various algorithm strengths are combined. This significantly reduces potential errors and minimizes variations noted with individual algorithm biases. Combined techniques include Random Forest, Gradient Boosting, and Decision Tree models (Brouwer et al., 2024). Furthermore, to ensure that the algorithm model remains generalizable despite variations noted in the training data, regularization techniques—such as Lasso and Ridge—are applied for complex algorithmic models (Avalos et al., 2020; El Alaoui et al., 2022).
Creating balanced data
To create a balanced dataset, AI/ML can instruct resampling techniques if some blood conditions are underrepresented (Avalos et al., 2020). Minority or majority classes of the dataset are either oversampled or undersampled to create balance. This contributes to improved model performance by analyzing various conditions. Algorithmic techniques such as Synthetic Minority Over-sampling Technique (SMOTE) can generate synthetic examples of underrepresented blood conditions, ensuring that the algorithm model learns from a broader and more diverse training dataset, which is representative of real-world cases (El Alaoui et al., 2022).
Engineering features
Complex data is managed through Principal Component Analysis (PCA) by ensuring the variance of a dataset and that the most relevant information of CBC data remains intact (El Alaoui et al., 2022). Integrating the measured CBC data from the analyzer with its associated clinical history provides insight into essential factors that could influence the data. This further enhances the algorithm’s ability to include variations applicable to the patient's clinical presentation.
Continuous learning models
Improvements in AI/ML models largely depend on new and updated data inputs to ensure continuous learning. The classification of blood components and potential diagnoses improves when new patterns develop and data variations occur over time. These techniques and methods collectively enhance the algorithm’s abilities, ensuring that accurate results are produced despite variability in clinical data (Avalos et al., 2020; El Alaoui et al., 2022).
Some of the training datasets include data for digital peripheral smear analysis, such as the Blood Cell Count dataset and ALL-IDB (Acute Lymphoblastic Leukemia Image Database); extensive numerical data from CBC analyses; AI-generated data in limited real-world settings for rare conditions; large collaborative datasets available through research institutes; and proprietary datasets from Sysmex and Beckman Coulter (Herman et al., 2021; Matek et al., 2021).
Practical integration potential of AI/ML for CBC analysis
Integrating AI/ML in instruments for CBC analysis has the potential to decrease dependence on manual peripheral smear reviews significantly. Several methods exist where AI and ML can be integrated:
Automated image analysis
As noted in other pathology disciplines, such as histology and cytology, digital microscopy is becoming favorable. AI/ML algorithms can analyze numerous digital images of blood smears with high accuracy, enabling the identification of different cell types and the detection of abnormalities. Furthermore, the need for manual reviews by technical staff members is reduced, streamlining the diagnostic process (Deshpande et al., 2021; Lin et al., 2023; Obstfeld, 2023; Yang et al., 2023).
Integrating CNNs with CBC analyzers and digital microscopy, deep learning models can be trained on reviewed datasets of blood smear images (Matek et al., 2021; Ben-Horin, 2024). These models can recognize different patterns associated with various hematological disorders. Cell types are classified, anomalies are detected, and abnormal cells are flagged for immediate attention (Gedefaw et al., 2023).
Enhancing diagnostic accuracy
Through CBC analysis, AI/ML methods can accurately predict conditions based on historical data. These methods identify patients potentially at high risk of developing specific conditions, ensuring that healthcare providers prioritize these cases. Implementing these AI/ML methods allows for effective resource allocation. Studies have also indicated that integrating AI/ML with conventional CBC analysis proved to match or exceed the level of interpretation in comparison with experienced hematologists (Alanzi et al., 2023).
Integrating CBC results with clinical data, such as a patient’s demographics, symptoms, and clinical history, enhances reliable diagnostic accuracy (Gedefaw et al., 2023). It reduces the need for unnecessary manual peripheral smear reviews while identifying high-probability cases, ensuring timely analysis, and improving informed decision-making. This is also deemed beneficial in emergency scenarios where an immediate response to critical conditions is essential (Alanzi et al., 2023).
Optimizing the workflow
Decisions made with integrated algorithms through AI/ML-driven support systems aid healthcare professionals in interpreting CBC results. This also assists in determining whether a peripheral smear review is, in fact, essential. Evidence-based recommendations from AI/ML reduce subjective interpretations that are likely to lead to a manual review of peripheral smears compared to conventional protocols (Avaloz et al., 2020; Zhu et al., 2022).
Applying different algorithms to CBC parameter evaluation also elevates its capability of early disease detection. AI/ML identify subtle changes that could otherwise be missed with a manual peripheral smear review. AI/ML methods improve their predictive capabilities over time by continuously learning from the availability of updated training datasets. This provides a current and up-to-date system that evolves with new medical knowledge and practices, further contributing to a lesser need for technical staff to intervene with manual methods (Gedefaw et al., 2023).
Healthcare professional training and exposure
The key to combining advanced technologies with traditional methods is ensuring the user experience is clearly understood (Herman et al., 2021; Gedefaw et al., 2023). Professional healthcare workers must be trained in AI/ML methodologies, optimize work processes, and effectively implement them in the clinical laboratory environment. Additionally, it is crucial to understand how AI/ML generates its insights and recommendations and interprets these findings effectively while simultaneously reducing labor-intensive and time-consuming tasks (Deshpande et al., 2021).
It enables healthcare professionals to make concise and confident decisions with increased accuracy and precision based on implementing new and improved techniques instead of the conventional peripheral smear examination. The outcomes are rapid diagnosis, early intervention, and prioritization of patient care (Deshpande et al., 2021). AI/ML-coupled CBC analysis provides a guideline for prioritizing critical patient cases.
Conclusion
Integrating AI/ML with existing hematology analyzers for CBC analysis emphasizes a significant technological advancement in hematology diagnostics. This evolutionary change in CBC analysis enhances diagnostic accuracy, automates repetitive tasks, and optimizes clinical laboratory workflows and processes. It further reduces turnaround time, the constant need for manual peripheral blood review, and the risk of human error.
AI/ML has the potential to transform traditional hematology practices, making patient care more efficient and effective. It provides strategies for early intervention and reduces unnecessary diagnostic test requests that are part of the conventional differential diagnosis approach. These technologies use advanced algorithms for real-time analysis and enhance pattern recognition and predictive analytics to assess complex cellular patterns and abnormal morphologies. Multimodal integration improves forecasting tools by combining a patient’s clinical history, demographics, and historical CBC data. The continuous learning capability of AI/ML ensures accurate diagnostics in the evolving healthcare environment.
Advanced algorithmic models are valuable tools for healthcare professionals. They provide useful insights and help reduce redundant, labor-intensive, and time-consuming tasks, allowing workers to focus on critical process improvements, research, and professional development activities.
Therefore, researchers, healthcare professionals, and technology innovators are encouraged to continue exploring the evolving field of AI/ML-integrated hematology diagnostics.
Prospective areas for future investigation
Future research should examine the effectiveness of synthetic data generation, utilizing advanced techniques such as Generative Adversarial Networks (GANs) and CNNs. Comparative studies that examine synthetic data generated by GANs alongside learning models using real-world data can reveal the potential for detecting rare hematology disorders. The main focus of this research is the ability of synthetic data creation to mimic real-world anomalies for accurate diagnostic hematology.
Conducting a longitudinal cohort study could highlight the benefits of using longitudinal datasets to evaluate patient history and diagnostic outcomes over time. This investigation can determine the accuracy of AI/ML predictive modeling tools, focusing on trend analysis and early diagnosis.
Investigate a multidisciplinary approach by combining numerical CBC data, digital imaging, clinical history, genetic testing, flow cytometry, and other additional test results into a unified AI/ML model. The purpose is to evaluate how accurately this multimodal approach contributes to an accurate diagnosis.
To determine the impact of AI/ML-enhanced CBC analyses, a retrospective study comparing this to traditional diagnostic methods will reveal the effects of measurement metrics, such as diagnostic accuracy, turnaround time, and healthcare professional satisfaction for future and conventional methods.
Using qualitative research methods, examine the ethical implications associated with AI/ML integration in the clinical laboratory setting by conducting interviews or surveys regarding data privacy, transparency, and the accuracy of laboratory results produced by AI/ML-driven diagnostics.
Glossary
Term | Definition |
Artificial Intelligence (AI) | AI mimics human intelligence by learning, reasoning, and self-correcting. It learns by accumulating data and its associated rules, reasons by applying those rules to datasets to reach conclusions and self-corrects by learning from past information and the rules it applies. |
Machine Learning (ML) | This is a subset of AI that can learn from information without being pre-programmed. It improves performance by continuously exposing itself to data rather than adhering to specific rules. |
Convolutional Neural Networks (CNNs) | are a deep learning technique explicitly designed to analyze images. They use multiple layers of filters to identify different morphologies of blood cells. This capability enhances rapid clinical decision-making and intervention through digital microscopy. |
Principal Component Analysis (PCA) | comprises statistical analysis to simplify complex data while retaining essential information, especially when datasets have a lot of components or features. |
Generative Adversarial Networks (GANs) | are an aspect of AI that generates new data that resembles existing data, especially when real-world data is rare or challenging to obtain. |
Table 1: A glossary of technical terms discussed throughout the article
References
1. Alanzi, T., Alanazi, F., Mashhour, B., Altalhi, R., Alghamdi, A., Al Shubbar, M., Alamro, S., Alshammari, M., Almusmili, L., Alanazi, L., Alzahrani, S., Alalouni, R., Alanzi, N., & Alsharifa, A. (2023). Surveying Hematologists' Perceptions and Readiness to Embrace Artificial Intelligence in Diagnosis and Treatment Decision-Making. Cureus, 15(11), e49462. https://doi.org/10.7759/cureus.49462
2. Avalos, M., Touchais, H., & Henriquez-Henriquez, M. (2020). Optimising Criteria for Manual Smear Review Following Automated Blood Count Analysis: A Machine Learning Approach. International Conference on Innovations in Bio-inspired Computing and Applications.
3. Ben-Horin, G. (2024, March 4). There is More to AI Than Meets the Eye | Scopio Labs. Scopio Labs. https://scopiolabs.com/ai/there-is-more-to-ai-than-meets-the-eye/
4. Brouwer, L., Cunney, R., & Drew, R. J. (2024). Predicting community acquired bloodstream infection in infants using full blood count parameters and C-reactive protein; a machine learning study. European journal of pediatrics, 183(7), 2983–2993. https://doi.org/10.1007/s00431-024-05441-6
5. Deshpande, N. M., Gite, S., & Aluvalu, R. (2021). A review of microscopic analysis of blood cells for disease detection with AI perspective. PeerJ. Computer science, 7, e460. https://doi.org/10.7717/peerj-cs.460
6. El Alaoui, Y., Elomri, A., Qaraqe, M., Padmanabhan, R., Yasin Taha, R., El Omri, H., El Omri, A., & Aboumarzouk, O. (2022). A Review of Artificial Intelligence Applications in Hematology Management: Current Practices and Future Prospects. Journal of medical Internet research, 24(7), e36490. https://doi.org/10.2196/36490
7. Freepik | Create great designs, faster. (n.d.). Freepik. http://www.freepik.com/
8. Gedefaw, L., Liu, C. F., Ip, R. K. L., Tse, H. F., Yeung, M. H. Y., Yip, S. P., & Huang, C. L. (2023). Artificial Intelligence-Assisted Diagnostic Cytology and Genomic Testing for Hematologic Disorders. Cells, 12(13), 1755. https://doi.org/10.3390/cells12131755
9. Haider, R. Z., Ujjan, I. U., Khan, N. A., Urrechaga, E., & Shamsi, T. S. (2022). Beyond the In-Practice CBC: The Research CBC Parameters-Driven Machine Learning Predictive Modeling for Early Differentiation among Leukemias. Diagnostics (Basel, Switzerland), 12(1), 138. https://doi.org/10.3390/diagnostics12010138
10. Herman, D. S., Rhoads, D. D., Schulz, W. L., & Durant, T. J. S. (2021). Artificial Intelligence and Mapping a New Direction in Laboratory Medicine: A Review. Clinical chemistry, 67(11), 1466–1482. https://doi.org/10.1093/clinchem/hvab165
11. Katare, D., Kourtellis, N., Park, S., Perino, D., Janssen, M., & Ding, A.Y. (2022). Bias Detection and Generalization in AI Algorithms on Edge for Autonomous Driving. 2022 IEEE/ACM 7th Symposium on Edge Computing (SEC), 342-348.
12. Lin, E., Fuda, F.S., Luu, H.S., Cox, A.M., Fang, F., Feng, J., & Chen, M. (2023). Digital pathology and artificial intelligence as the next chapter in diagnostic hematopathology. Seminars in diagnostic pathology.
13. Makarov, N., Bordukova, M., Rodriguez-Esteban, R., Schmich, F., & Menden, M. P. (2024). Large Language Models forecast Patient Health Trajectories enabling Digital Twins. MedRxiv (Cold Spring Harbor Laboratory). https://doi.org/10.1101/2024.07.05.24309957
14. Matek, C., Krappe, S., Münzenmayer, C., Haferlach, T., & Marr, C. (2021). Highly accurate differentiation of bone marrow cell morphologies using deep neural networks on a large image data set. Blood, 138, 1917 - 1927.
15. Meiseles, A., Paley, D., Ziv, M., Hadid, Y., Rokach, L., & Tadmor, T. (2022). Explainable machine learning for chronic lymphocytic leukemia treatment prediction using only inexpensive tests. Computers in biology and medicine, 145, 105490. https://doi.org/10.1016/j.compbiomed.2022.105490
16. Mutema, L., Chapanduka, Z., Musaigwa, F., & Mashigo, N. (2021). In-depth investigation of turn-around time of full blood count tests requested from a clinical haematology outpatient department in Cape Town, South Africa. African journal of laboratory medicine, 10(1), 1318. https://doi.org/10.4102/ajlm.v10i1.1318
17. Obstfeld A. E. (2023). Hematology and Machine Learning. The journal of applied laboratory medicine, 8(1), 129–144. https://doi.org/10.1093/jalm/jfac108
18. Padmanabhan, R., El Alaoui, Y., Elomri, A., Qaraqe, M. K., El Omri, H., & Yasin Taha, R. (2023). Machine Learning for Diagnosis and Screening of Chronic Lymphocytic Leukemia Using Routine Complete Blood Count (CBC) Results. Studies in health technology and informatics, 305, 279–282. https://doi.org/10.3233/SHTI230483
19. Srisuwananukorn, A., Salama, M. E., & Pearson, A. T. (2023). Deep learning applications in visual data for benign and malignant hematological conditions: a systematic review and visual glossary. Haematologica, 108(8), 1993–2010. https://doi.org/10.3324/haematol.2021.280209
20. Wencke Walter, Christian Pohlkamp, Manja Meggendorfer, Niroshan Nadarajah, Wolfgang Kern, Claudia Haferlach, Torsten Haferlach. Artificial intelligence in hematological diagnostics: Game changer or gadget? Blood Reviews, Volume 58, 2023, 101019, ISSN 0268-960X, https://doi.org/10.1016/j.blre.2022.101019.
21. Yang, R., Chen, D., Yang, Q., Qiu, Y., & Wang, F. (2024). Accurate COVID-19 detection using complete blood count data and machine learning. International Journal of Metrology and Quality Engineering.
22. Yang, Y., Guan, S., Ou, Z., Li, W., Yan, L., & Situ, B. (2023). Advances in AI‐based cancer cytopathology. Interdisciplinary Medicine, 1.
23. Zhu, J., Lemaire, P., Mathis, S., Ronez, E., Clauser, S., Jondeau, K., Fenaux, P., Adès, L., & Bardet, V. (2022). Machine learning-based improvement of MDS-CBC score brings platelets into the limelight to optimize smear review in the hematology laboratory. BMC cancer, 22(1), 972. https://doi.org/10.1186/s12885-022-10059-8
About Renaldo Pool, BHSc
As a medical laboratory scientist, I'm passionate about scientific research and writing. I combine theory and practice to explore healthcare advancements. My lab expertise helps me investigate areas for improvement in healthcare through research and practical implementation. I aim to conduct thorough studies to advance medical knowledge and aid healthcare professionals in decision-making. Ultimately, I strive to bridge the gap between scientific research and practical application, contributing positively to the general population's health.
