In Africa, the integration of AI into haematology presents significant opportunities to address persistent challenges related to healthcare infrastructure and workforce limitations.¹⁴ Laboratories in several African countries, including Nigeria, Kenya, and South Africa, often face inadequate infrastructure, shortages of trained personnel, and limited resources. These constraints frequently result in prolonged diagnostic turnaround times and variability in test accuracy and consistency.^15,16 Despite these challenges, the current landscape provides a strong foundation for AI-driven innovation, offering scalable solutions that can enhance efficiency and accessibility in diagnostic services.¹⁴

Artificial intelligence-powered digital morphology platforms and image interpretation systems have the potential to decentralise diagnostic capabilities across the continent. By enabling smaller or remote laboratories to access centralised or cloud-based AI platforms, these technologies can improve diagnostic coverage while promoting standardisation of results.¹⁷ Pilot initiatives in countries such as Kenya and South Africa are already exploring the integration of AI into pathology workflows, with the aim of enhancing diagnostic accessibility and improving consistency in reporting.¹⁷ Furthermore, the United States Agency for International Development has highlighted AI’s potential to ‘leapfrog’ conventional diagnostic pathways by allowing machine learning (ML) algorithms to screen digitised slides, identify regions of interest for clinician review, or, in some cases, directly support diagnostic decision-making through pattern recognition.¹⁷

The deployment of AI-based tools for the automated detection of red blood cell–related diseases has also been reported in West Africa.¹⁸ These systems employ advanced ML techniques to analyse red blood cell morphology and identify abnormalities associated with a range of haematological conditions. Studies integrating artificial neural networks and convolutional neural networks (CNNs) with traditional classifiers such as support vector machines and k-nearest neighbour algorithms have demonstrated enhanced diagnostic performance, with reported accuracies ranging from 91% to 98%. Collectively, these developments underscore the dual role of AI as both a diagnostic support mechanism and a capacity-building tool within African healthcare systems.¹⁶

This narrative review critically examines current applications of AI in haematology laboratory diagnosis across key thematic domains, morphological diagnostics, flow cytometry, cytogenetics and genomics, and CDS. It also addresses ethical, regulatory, and economic considerations relevant to AI implementation, with particular attention to resource-limited contexts. Finally, the review provides future perspectives for sustainable and equitable integration of AI in global and African haematology practice.

The literature in this field has expanded rapidly in recent years. However, available evidence is scattered across studies with different designs and methodologies. This fragmentation limits clear interpretation and practical application. In addition, findings are sometimes inconsistent and lack unified synthesis. A narrative review is therefore needed to integrate current knowledge, clarify key concepts, and highlight unresolved issues within the field.

This narrative review aims to provide a focused and critical synthesis of the existing literature. It summarises key molecular, diagnostic, and translational findings. It compares results across different study types. It also identifies limitations, knowledge gaps, and priorities for future research.

The application of AI in morphological haematology has transformed how blood and bone marrow samples are analysed. Artificial intelligence-driven image analysis systems enable objective, high-throughput, and reproducible cellular classification, minimising human bias. Convolutional neural networks have demonstrated remarkable performance in identifying leukocyte subtypes and detecting abnormal morphologies.^19,20,21 Widely adopted platforms such as CellaVision and Morphogo have achieved classification accuracies of 95% – 98% for standard leukocyte types, matching and sometimes surpassing manual microscopy.^{20,21,22,23,24,25,26,27} These systems not only enhance diagnostic accuracy but also reduce inter-observer variability and optimise technologists’ time by automating routine differential counts.

More advanced hybrid models that integrate CNNs with transformer architectures have further improved recognition of subtle morphological abnormalities. These hybrid systems can detect dysplastic changes characteristic of myelodysplastic syndromes and identify abnormal red cell morphologies such as schistocytes and spherocytes.^28,29,30,31 Through continuous learning on large annotated datasets, these algorithms capture minute, non-obvious features that may be challenging for humans to interpret consistently, thereby supporting more standardised and reproducible reporting.

Incorporating whole-slide imaging has expanded the utility of AI in digital haematopathology. Modern systems can digitise glass slides into high-resolution images for AI-assisted analysis and pathologist review.^32,33 This integration enables rapid detection, quantification, and prioritisation of abnormal regions, facilitating both diagnostic accuracy and workflow efficiency. Moreover, cloud-based AI platforms now enable remote consultation and collaborative interpretation, an important advancement for laboratories across Africa, where access to haematopathologists may be limited. By enabling virtual slide review and automated differential analysis, AI-powered whole-slide imaging platforms can bridge geographical gaps and contribute to equitable diagnostic access.

While AI-based morphology systems demonstrate high diagnostic accuracy across multiple studies, most remain confined to controlled research environments. Broader multicentre validation and cost-effectiveness evaluations are still required before widespread clinical deployment, particularly in resource-limited settings.

The integration of AI and ML into flow cytometry has revolutionised how complex, high-dimensional data are analysed. Traditionally, manual gating, a subjective, time-consuming process, relied heavily on expert interpretation, often introducing human bias. Artificial intelligence now enables automated, reproducible, and efficient analysis through advanced clustering, visualisation, and classification techniques.

Unsupervised algorithms such as FlowSOM and t-SNE allow for unbiased clustering of cells based solely on marker expression, effectively reducing analyst-to-analyst variability.^34,35 These tools automate data organisation and visualisation, providing objective identification of cellular subpopulations. Building on this foundation, supervised deep learning models such as DeepFlow learn from expert-defined cell populations and apply this knowledge to new datasets, further standardising results and reducing turnaround time.

Beyond automation, ML has enhanced diagnostic insight: ML models have shown superior performance in detecting rare cell populations and subtle immunophenotypic changes associated with haematological malignancies.^36,37,38 By integrating multiple parameters simultaneously, these models provide sensitive and quantitative assessments of disease status particularly for detecting minimal residual disease in acute myeloid leukaemia and myelodysplastic syndromes.^36,38 Artificial intelligence-based minimal residual disease analysis reduces subjectivity and provides robust, reproducible results, which is critical for both initial diagnosis and post-treatment monitoring.

In the African context, AI-enabled flow cytometry offers significant potential to improve access to standardised diagnostic testing. Centralised AI-assisted cytometry platforms could support smaller or under-resourced laboratories by processing uploaded data through secure, cloud-based systems. This approach would allow regional centres, such as those in Nigeria, South Africa, and Kenya, to share computational resources while maintaining local data governance. Such collaborative networks could reduce the current gap in advanced haematology diagnostics across the continent and enhance training capacity for laboratory scientists.

Despite notable success in automating complex flow cytometry data interpretation, algorithmic transparency and clinical validation across diverse populations remain limited. Integration into African laboratory workflows will depend on infrastructure investment and standardised regulatory oversight.

Artificial intelligence applications in cytogenetics and genomics have advanced rapidly, automating complex image interpretation and data analysis tasks once reliant on manual expertise. Deep learning algorithms are now capable of identifying chromosomal abnormalities, predicting pathogenic variants, and integrating multi-omic data to inform diagnosis and prognosis.

In cytogenetics, CNNs have been developed to automate chromosome segmentation and classification. Tools such as KaryoNet and Varifocal-Net can accurately localise metaphase spreads, isolate individual chromosomes, and identify subtle banding pattern abnormalities.^{39,40,41,42,43} These AI-based systems demonstrate accuracy levels exceeding 98% in classifying chromosomal structures, significantly reducing human workload and inter-observer variability. Moreover, they facilitate a human-in-the-loop workflow, allowing cytogeneticists to focus on verifying AI-flagged anomalies, such as deletions, translocations, and duplications.

Artificial intelligence has also expanded into genomic interpretation, particularly for analysing next-generation sequencing data. Models such as AlphaMissense and MissenseNet predict the pathogenicity of missense variants with high accuracy,^44,45,46 while frameworks like MethylNet and ChINN enable the interpretation of DNA methylation and chromatin interaction data^47,48 (Table 1). These developments are transforming genomics from descriptive to predictive science, enabling clinicians to infer biological function and clinical impact directly from sequence data.^49,50,51

In resource-limited settings, including many parts of Africa, the use of AI in cytogenetics and genomics remains limited but promising. Cloud-based AI genomics platforms can offer affordable, scalable tools for laboratories without advanced sequencing infrastructure and preserve data privacy while pooling genomic data across African borders. This strategy could help address the chronic shortage of large, representative datasets required for model validation and ensure that African genomic diversity is adequately reflected in AI-driven analyses (Figure 1).

Current AI-based cytogenetic and genomic tools show strong computational performance but are largely in prototype or academic stages. The translation of these models into clinical diagnostics will require harmonised data governance frameworks and equitable access to genomic resources across low- and middle-income regions.

Artificial intelligence is playing an increasingly important role in CDS within haematology. Beyond diagnostic applications, AI models are now being used to assist in therapeutic decision-making, risk stratification, and patient monitoring. By integrating laboratory findings, genomic data, and clinical variables, AI-based CDS systems can enhance personalised treatment approaches and improve workflow efficiency.

Ensemble learning models, for example, have been applied to risk stratification in acute myeloid leukaemia, combining ML outputs with established clinical guidelines to support treatment planning.^52,53 Reinforcement learning algorithms have further demonstrated potential in optimising dynamic treatment strategies, such as chemotherapy dosing and transfusion management.^54,55,56 These systems continuously learn from patient data and treatment outcomes, enabling adaptive protocols tailored to individual responses.

Such technologies complement traditional decision-making rather than replace it. By providing evidence-based recommendations and predictive insights, AI-driven CDS platforms allow clinicians to focus on complex cases that require expert interpretation. Importantly, this synergy between clinician expertise and AI analytics enhances both accuracy and efficiency in patient care.

Although decision-support algorithms have demonstrated utility in diagnostic prediction and therapy selection, their ethical governance and cost-effectiveness in routine haematology practice are yet to be established. Regional pilot programmes and transparent validation protocols are essential next steps.

The rapid integration of AI into haematology laboratories raises important ethical, regulatory, and health-economic questions that extend beyond technical performance and must be addressed to ensure responsible and equitable implementation. While AI-driven systems demonstrate high accuracy and efficiency, their clinical impact depends on trust, governance, and accessibility within real-world healthcare settings.

Although AI has already demonstrated substantial promise in haematology diagnostics, several challenges must be addressed before widespread and sustainable clinical adoption can be achieved. Future development should prioritise explainability, generalisability, and ethical accountability to ensure patient safety and clinician acceptance.

Artificial intelligence has emerged as a transformative force in haematology laboratory diagnostics, offering measurable improvements in accuracy, reproducibility, and efficiency across morphological assessment, flow cytometry, cytogenetics, genomics, and CDS. These advances signal a shift toward data-driven and personalised haematology practice.

Despite significant progress, widespread clinical adoption remains constrained by challenges related to external validation, model explainability, regulatory uncertainty, and unequal access to digital infrastructure. Ethical oversight and economic considerations must therefore guide AI deployment to ensure that technological innovation aligns with equitable healthcare delivery.

In Africa, AI holds exceptional potential to strengthen diagnostic capacity, expand access to advanced haematology services, and promote regional self-sufficiency in laboratory medicine. Through collaborative networks, federated learning frameworks, and context-sensitive policy development, African laboratories can actively contribute to and benefit from global AI advancement.

Ultimately, the successful integration of AI into haematology will depend on sustained collaboration among clinicians, data scientists, policymakers, and technology developers. By fostering transparent, evidence-based, and ethically grounded AI systems, the field can realise the full potential of AI as a trusted partner in haematology diagnostics, improving patient outcomes on both regional and global scales.