A narrative review on artificial intelligence in haematological malignancies: from diagnostic precision to therapeutic innovation

Introduction

Haematological disorders, such as leukaemia, lymphoma, myelodysplastic syndrome (MDS), multiple myeloma (MM), and myeloproliferative neoplasms (MPNs), are heterogeneous diseases, making diagnosis and treatment challenging. Routine diagnostic approaches include morphological diagnosis, identification of immunological markers, cytogenetic evaluation, and molecular biology techniques, which require a high level of clinical expertise and specialised equipment, often inaccessible in under-resourced areas, leading to variations in healthcare delivery (1).

Additionally, many haematological malignancies are resistant to conventional treatments, including supportive therapy, radiotherapy, and immunotherapy, which are tailored to specific symptoms and disease types. Although novel treatments such as targeted therapy, immunotherapy, and hematopoietic stem cell transplantation (HSCT) have emerged, patients may still face relapse, drug resistance, or transplant-related complications (2,3).

Thus, applying artificial intelligence (AI) offers a sophisticated technological approach that mimics human reasoning. In 1995, it was first utilised in the haematology field for peripheral blood (PB) smear and bone marrow (BM) screening, and it became widely applied due to its high accuracy and specificity (4).

Machine learning (ML) is a subtype of AI that enables systems to learn from data without explicit programming (5,6). Deep learning (DL) is a subtype of ML that extracts information from multiple input data sources, such as images and numbers (7). DL has several neural network architectures recently used in various clinical applications, including artificial neural networks (ANNs) and their advanced forms, deep neural networks (DNNs), recurrent neural networks (RNNs), and convolutional neural networks (CNNs) (8). Several haematological diseases can be diagnosed using a CNN-based image-processing model (9,10).

Besides DL, other ML algorithms, such as support vector machines (SVMs), random forests (RFs), genetic algorithms (GAs), and natural language processing (NLP), are widely used to enable computers to process the massive amounts of medical data efficiently (11). These tools are applied to morphological analysis, immunophenotyping, chromosome karyotyping, gene sequencing, biomarker discovery, drug development, risk assessment, classification of leukaemia, and prognosis monitoring (10,12,13).

AI has the ability to analyse large datasets at relatively low marginal cost per case, making AI ideal for developing dynamic, remote-access diagnostic models to enhance treatment and personalise therapy decisions (14).

United States Food and Drug Administration (US FDA) -approved tools, such as CellaVision, Scopio LabsX100, Techcyte, and Morphogo, utilise ML to automate cell morphology analysis in BM and blood smears, while DeepFlow has streamlined immunophenotyping via flow cytometry (FC) (15-19). In cytogenetics, CNN-based tools such as Varifocal-Net and KaryoNet have improved low-resolution karyotyping and the detection of cryptic chromosomal abnormalities, as seen in Figure 1 (20,21).

Figure 1 Role of AI in the comprehensive diagnosis of haematological malignancies. AI, artificial intelligence.

In genomics, DL combined with high-throughput sequencing has uncovered disease heterogeneity. For instance, the chromatin interaction neural network (ChINN) overcomes the drawbacks of regional genomic assessments (22).

ML also plays an important role in personalised haematology care. It enables data-driven prediction models, risk stratification, and individualised treatment planning based on patient-specific clinical data (23,24). Furthermore, ML aids in identifying new therapeutic targets, supporting innovation in haematological treatments (25).

Despite these benefits, AI implementation in haematology still faces multiple barriers, such as a lack of product certification standards, inconsistent data formats, limited interdisciplinary collaboration, ethical concerns, and the fuzzy nature of many AI models (“black box” algorithms). Addressing these issues requires regulatory oversight, technical improvement, and enhanced training for both clinical haematologists in AI and data science (26).

There is an urgent need for efficient, precise, and economical diagnostic and treatment tools for haematological malignancies. These accelerated the exploration of AI-assisted solutions. As data volumes grow and AI technology advances, the integration of AI into medical applications, particularly in haematology, has gained momentum.

This review explores the applications of AI in haematology in recent years, focusing on diagnostic applications including morphology, immunological markers, cytogenetics, and molecular assessments. It highlights AI’s role in disease diagnosis, therapeutic invention, disease risk evaluation, and prognosis monitoring. Looking ahead, investigating the combination of AI with haematological tools could potentially enhance the intelligence, standardisation, and consistency of haematology practices.

It is worth mentioning that this review specifically addresses the application of AI tools in haematological malignancies such as leukaemia and lymphoma, covering the period from 2017 to 2024, providing a more detailed and up-to-date evaluation of diagnostic, prognostic, and therapeutic applications. This article is presented in accordance with the Narrative Review reporting checklist (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-195/rc).

Methods

This review has been designed to investigate the role of AI in diagnosing, treating, and prognosing haematological malignancies by reviewing all relevant articles published from 2017 to 2024. The search of this review is based on related keywords in PubMed, Scopus, Google Scholar, and Web of Science (WOS) databases. The main keywords used for searching in this review were “haematological malignancies”, “artificial intelligence”, “diagnosis”, “treatment”, and “prognosis”. The searching has been classified according to the role of AI in the diagnosis of hematologic malignancies, the role of AI in the treatment of haematological malignancies, and the role of AI in predicting the risk of haematological malignancies.

We extracted specific information from the research articles, including current advancements in AI tools for improving the diagnosis and treatment of haematological malignancies, as well as the challenges these tools face in promoting the standardisation of clinical diagnosis and treatment in the haematological field.

The criteria for selecting articles for this review were based on original research articles written in English that provided relevant information. However, several exclusion criteria were applied, including articles published in languages other than English, review articles, conference papers, book chapters, and unrelated articles. The process of selection and identification of the articles has been explained in a flow diagram, as seen in Table S1 and Figure 2.

Figure 2 The search methodology workflow for this narrative review. WOS, Web of Science.

Applications of AI in haematological malignancies

Role of AI in the diagnostic evaluation of haematological malignancies

The application of AI in the morphological diagnosis of haematological malignancies

Morphological diagnosis of BM and PB is the most common and effective diagnostic procedure for haematological malignancies. Traditional methods of morphological diagnosis involve assessing the size, shape, and structure of blood cells and identifying the types and classifications of haematological malignancies such as leukaemia (27,28). Therefore, accurate classification of leukaemia subtypes, such as acute lymphoblastic leukaemia (ALL), acute myeloid leukaemia (AML), chronic lymphocytic leukaemia (CLL), and chronic myeloid leukaemia (CML), is crucial for designing the appropriate treatment plan (29). Microscopic diagnosis has relied heavily on the experience of haematologists to examine samples under the microscope. However, this process is prolonged, impractical, and tends to inter-observer variability (30). Thus, AI systems have emerged as a powerful tool in revolutionising medical diagnostics by enhancing the accuracy, efficiency, and reliability of morphological diagnosis, as seen in Table 1. For example, one study applied ML for the detection and classification of AML cells using the RF algorithm to identify the segment of nucleus and cytoplasm based on morphological features from 100 AML patients and 100 healthy controls with a total of 18,365 images. They have achieved 92.99% accuracy in identifying sixteen features for each image of white blood cells (WBCs) from the blood smear sample. Furthermore, they have achieved 93.45% accuracy for identifying immature cells such as erythroblasts, monoblasts, promyelocytes, and myeloblasts (31).

On the other hand, another has developed and trained an AI model called a mask region-based CNN (Mask R-CNN)—which was specifically designed to detect and classify nucleated cells in BM aspiration smears, thereby addressing the far more complex challenge of BM cytology versus PB. They trained a Mask R-CNN model on a large data set of 542 BM slides, and the model was able to distinguish 15 cell categories with an accuracy of 0.94 on a dedicated 26,170 dataset. Interestingly, the model reached a classification accuracy of 0.881. The cell type percentage differentials produced by the algorithm of Mask R-CNN showed a strong correlation (ρ>0.8) with the manual counts method for most cell categories. This work represents a promising count method for BM cells that could reduce the technically challenging task of BM analysis. Interestingly, the Taiwan Food and Drug Administration approved this Mask R-CNN as an AI model in 2021 for BM analysis (58).

Target detection models such as the You Only Look Once (YOLO) model and its recent versions (YOLOv8 and YOLOv11) have been developed to precisely recognise and categorise each target cell in BM smears (59). The latest study has utilised different models, namely YOLOv8, YOLOv11, residual network with 50 layers (ResNet50), and inception residual network-version 2 (ResNet-v2), with image processing for the detection of ALL from a dataset with high accuracy rates of 99.7% (32,60).

CNN model, a class of DL algorithms trained on labelled BM smear datasets to recognise and classify various types of WBCs and to identify malignant or atypical cells in stained smear images. For example, one study has been able to differentiate between AML, MDS, and aplastic anaemia and investigate MDS subtypes with an accuracy of 92.9% when applying the CNN method to various BM smear pictures (33).

Another study has combined CNNs with other ML algorithms, including SVM and K-nearest neighbour (KNN), to classify 250 images of BM smears. First, to distinguish WBCS from red blood cells (RBCs) and platelets, the KNN algorithm was applied, followed by the watershed method to isolate overlapped cells. Then, the SVM algorithm was applied to the images for the classification of leukaemia. Interestingly, the combination of these two algorithms achieved 97.5% accuracy in leukaemia classification and 94.75% accuracy in counting lymphocytes and monocytes (12).

Another study has used the SVM algorithm in combination with AlexNet to improve the diagnosis of ALL. AlexNet can detect the lymphocytes by extracting features of them, and SVM classifies and identifies the cells into normal lymphocytes and malignant lymphocytes. Interestingly, this approach has reached a 98% accuracy rate in detecting ALL from blood smear samples (34).

DL models have been innovated to accurately predict morphological features of leukaemia and predict common mutations in haematological malignancies from BM smears, such as the NPM1 mutation in AML (35). For example, a study of Eckardt et al. [2022] has utilised DL model to identify the most common mutation in AML patients (NPM1 mutation) from the BM slide. The model has identified NPM1 mutation based on specific morphology of NPM1 blast cells; cup-like blast, condensed chromatin and prominent nucleoli of mutant myeloblast. This approach could accelerate diagnostic workflows, but still a need for multicentre validation before approving for routine clinical use (38). Another study has applied DL pipeline to predict therapy-relevant genetic mutation in AML patients directly from the BM smear stained with Pappenheim. The study included over two million single images from 408 patients, and the results were validated with a cohort of 71 patients. This model has successfully predicted the essential mutant genes such as NPM1 and FLT3 mutations, CBFB:MYH11 fusions, and myelodysplasia-related cytogenetics with high accuracy. The model has worked in two stages: first, the cells were classified, and then a CNN was applied to detect mutations. However, the study was conducted in a single centre with uniform staining and diagnosis protocols (39).

In addition, the study of Wei et al. [2025] has represented the capability of DL based on multi-instance learning (MIL) in predicting NPM1 mutation and FLT3-ITD mutation in AML patients by analysing whole slide images. This model has improved the accuracy of predicting NPM1 mutation morphologically, but a potential improvement is needed to predict FLT3-ITD mutation (35).

Interestingly, different ML models such as RF, KNN, and SVM have been able to accurately identify the changes in biomarkers and risk stratification among paediatric patients with ALL and AML (36).

Regarding the detailed diagnosis of leukaemia, one study has applied the CNN model for the diagnosis of histopathological images of CML. The researchers were able to determine the phases of CML progression. Identifying the specific phase of CML a patient is in is crucial in determining the most appropriate treatment approach. The findings illustrated the ability of CNN in predicting several CML stages with 99.6% accuracy (37).

AI-based digital microscope imaging systems, such as Scorpio Labs X100, allow full-field PB smear investigations and remote viewing. These developments allowed earlier detection, improved accuracy, and reduced diagnostic subjectivity in morphological assessments (16,61).

Application of AI in the immunophenotypic diagnosis of haematological malignancies

Immunophenotyping uses monoclonal antibodies tagged with fluorochromes to classify leukaemia subtypes such as ALL, AML, CML, and CLL through FC and multiparameter FC (MFC) (62). The traditional FC analysis relies on manual result interpretation, which leads to time-consuming processes and potential human errors (62). Integrating AI with the FC method significantly advances the diagnosis with high accuracy and efficiency (Table 1). For instance, DL can identify the relationship between immunomarker patterns and molecular abnormalities in leukaemia, assisting clinicians with diverse differential diagnoses (63,64). Another study has used the Gaussian mixture model-SVM (GMM-SVM) as an advanced model of ML to identify and characterise the diagnostic cell populations and their phenotypic heterogeneity among AML samples (40).

DeepFlow, combined with a multidimensional density-phenotype coupling (MDPC) algorithm, has been used to diagnose and classify lymphocytic leukaemia (19,41). Another recent study has trained ResNet50 and EverFlow on FC data and has achieved a sensitivity of 94.6% in detecting AML patients and 98.2% sensitivity for B-ALL patients (42).

The SVM has been combined with FC to differentiate malignant from normal CML neutrophils using a four-colour detection panel (CD45, CD65s, CD15, and CD11b). They successfully identified mature neutrophils from malignant neutrophils with 95.80% sensitivity and 95.30% specificity (43).

Interestingly, another study has applied FC to AML patients to identify measurable residual disease (MRD). It has used clustering algorithms and ML for standardised multidimensional MRD analysis in AML patients through FC analysis. This study evaluated BM samples from 136 pre-relapse and 155 sustained remission controls. They developed a FC analysis pipeline called C-flow-MRD, which can report MRD with 74% sensitivity and almost 87% specificity in the cohort. However, these achievements require validation using larger datasets. In addition, this assay has other limitations, including a restricted number of markers and being limited to blasts expressing CD34 and/or CD117 (44).

Taken together, these promising findings illustrate the potential utility of AI-derived predictive models in highly accurate and more consistent diagnosis of haematological cancer.

Application of AI in cytogenetic karyotyping diagnosis of haematological malignancies

Cytogenetics is another diagnostic method for haematological malignancies that relies on analysing and interpreting the morphological features of the chromosomes. Haematological malignancies exhibit several chromosomal aberrations, such as translocations, deletions, inversions, and duplications. Identifying these abnormalities is fundamental for categorising patients into disease subtypes with accurate prognosis (65).

Karyotyping analysis has long been used in haematology and is one of the most important diagnostic tools for identifying a wide range of cytogenetic abnormalities among patients (66). Typically, haematologists identify chromosomes manually by conducting G-banded metaphase maps, which rely on the specific length and banding pattern of each chromosome. However, this method is time-consuming, expensive, and inefficient (66).

Various AI systems for automated and semi-automated karyotyping have been introduced to improve chromosome classifications and karyotyping. These AI-driven systems are increasing efficiency and accuracy, reducing the time and costs associated with manual chromosomal karyotyping (67,68). Interestingly, ML methods have demonstrated powerful potential for analysing large amounts of cytogenetic data, as seen in Table 1.

For example, in 2019, CNN models were integrated into the Varifocal-Net system and achieved a high accuracy (99.2%) in identifying chromosome types and polarity tasks (20). However, in 2022, the CNN model corrected rotation and positioning errors in the traditional karyotyping procedure with 98.8% accuracy (69). Additionally, several AI models have been used to address low-quality or poor-resolution karyotype maps by reconstructing high-resolution images, enabling the detection of abnormalities (45,46).

Recently, in 2023, DL-based approaches such as KaryoNet, masked feature interaction module (MFIM), and deep assignment module (DAM) were developed. KaryoNet has been utilised as a backbone to extract features of chromosomes based on R-band and G-band of chromosomes, and MFIM was applied to capture attention-based mechanisms to compare all the chromosomes, while DAM used combinatorial logic to assign each chromosome to its correct label. Taken together, DL-based approaches have improved the accuracy of normal chromosome classification by training it to characterise chromosomes according to R- and G-band. This provides a new method for accurately analysing the karyotype of patients with various types of numerical abnormalities. However, to ensure model accuracy in classifying complex structural chromosomal abnormalities, further validation is required before being integrated into diagnostic practice (21).

Application of AI in molecular-based diagnosis of haematological malignancies

Molecular biology tests are essential methods for the clinical diagnosis of haematological diseases, including polymerase chain reaction (PCR), DNA sequencing, single-cell RNA sequencing (RNA-seq), and DNA chip technology. These techniques are fundamental for diagnosing the progression of haematological diseases by screening various gene mutations in leukaemia cells, particularly those that may lead to the onset or relapse of the disease (70).

These advanced molecular technologies are commonly used for diagnosing haematological disorders, assessing disease prognosis, and planning treatment strategies for individual patients (71). Thus, AI model applications could hold transformative potential in molecular-based leukaemia diagnosis by enhancing accuracy, speed, and depth of analysis.

For example, ML has been applied to whole-genome sequencing to classify AML with high precision using blood gene expression profiles, aiding risk prediction and subtype classification. A combined ML tool with a transcriptome can improve genome-based diagnosis at low marginal cost. However, this study has collected data from several studies with different designs and goals. Therefore, the differences could affect the performance of the ML tool (47). SVM model accurately classifies AML based on known genes, while RF algorithms identify diagnostic genes in CML (48,49). One-class classification SVMs (OCSVM) method has been applied to 1153 normal datasets and 2,674 AML datasets from National Centre for Biotechnology Information (NCBI) profiles to identify the diagnostic mutant genes in AML. The results have shown that OCSVM is more efficient at predicting gene mutations in AML than other methods (48). Another study utilised the SVM tool in combination with the RF algorithm and identified gene expression, signalling pathways, and immune cell infiltration in 76 CML patients and 74 normal samples. However, the sample size would be the major limitation of this study to validate the performance of this model (49).

One of the most common causes of the pathogenesis of haematological malignancies, such as AML and ALL, is the hypermethylation of suppressor genes and hypomethylation of oncogenes, which may subsequently lead to a poor prognosis (72,73). Thus, the integration of various AI algorithms, such as DL, has had a significant impact on the discovery and analysis of DNA methylation at genomic targets (50).

Several studies have applied DL algorithms to develop EPISCORE. This single-cell histology reference model is based on analysing DNA methylation in cell heterogeneity using high-resolution datasets, which leads to reducing the cost of inferring DNA methylation differences in ALL patients (74,75). Another DL model, a modular deep learning framework for DNA methylation analysis (MethylNet), aims to discover unknown DNA methylation by combining unsupervised generation, clustering, cell-type deconvolution, subtype classification, age regression, and smoking status classification (51).

On the other hand, chromatin structure plays a role in diseases where alterations in chromatin structure have been shown to have pathogenic effects and coincide with the occurrence of leukaemia (76). ChINN, an AI tool, has been developed to recognise chromatin interactions of prognostic genes and to identify differential expression of oncogenes in samples from CLL patients (23).

Furthermore, next-generation sequencing (NGS) has been widely utilised as a molecular tool for accurate diagnosis and appropriate treatment for patients with haematological malignancies (77). Consequently, several attempts have been made to develop AI systems based on NLP algorithms for clinical sequencing applications, aiming at precision treatment for patients with haematological malignancies. For instance, one study has combined an NLP algorithm with NGS to identify mutations and associated pathways in malignant and normal tissues from over 150 patients with AML and ALL, thereby inferring applicable drug information (52). Another study has employed a similar algorithm and successfully sequenced and identified mutations from more than 300 patients with haematological cancers as input (53).

Taken together, AI demonstrates the potential to outperform experienced haematologists in the comprehensive analysis of high-dimensional whole-genome sequencing and NGS data, effectively narrowing interobserver variability and improving diagnostic timeliness and accuracy.

Lastly, RNA-seq is a crucial method for transcriptomic analysis, helping identify biological processes and the molecular mechanisms underlying the pathogenesis of haematological disorders (78). Today, with the development of RNA databases, it is possible to investigate the underlying mechanisms of blood cancer heterogeneity and the biological behaviours of cancer to identify new potential targets in the clinical setting (79,80). In transcriptomics, batch effects are easily generated because RNA-seq is particularly sensitive. Therefore, RCA2 and BERMUDA models have been developed to reduce batch effects while unifying single-cell data (54,55).

Application of AI in proteomics analysis for haematological malignancies diagnosis

Proteomics can address the limitations of genomic methods and enhance the discovery of biomarkers and personalised treatments for various leukaemias. Detecting protein variants from multivariate data requires preprocessing for the diagnosis of blood disorders (81). Thus, ML models have been applied in proteomics analysis due to their powerful ability to categorise unknown samples (82). In a recent study, an ML algorithm confirmed the disruption of the coagulation cascade response in acute promyelocytic leukaemia by identifying a new fusion protein target that regulates various aspects of the transcriptional response (56). Another study among AML patients has confirmed that ML can help to identify key protein pathways in FLT3 internal tandem duplication mutations (57). Overall, with the increasing publicity of proteomics and the generation of massive amounts of proteomic data, the future challenge depends on utilising AI to process this enormous amount of data effectively.

Role of AI in diagnostic support and data integration

Another important role of AI is facilitating data integration by unifying electronic health records (EHRs), devices, pathology reports, and population databases. Synthesising this varied information into a unified clinical picture enhances personalised decision-making, risk stratification, and overall healthcare. For example, AI-driven NLP can effectively extract data from unstructured clinical notes. In this way, comprehensive datasets can be developed to enhance the accuracy of diagnosis and research (83).

A recent study has identified how NLP creation of large language models, such as ChatGPT-4o, Claude 3 Opus, Meta Large Language Model A (MetaLLaMA 3), and Gemini, has been applied to 19^th-century paediatric unclassified leukaemia case reports. The NLP was essential in enabling these models to interpret unstructured, archaic medical text and extract relevant clinical features that are assessed in modern differential diagnosis. Intriguingly, three out of four of these models have been able to correctly diagnose CLL based on morphological clues such as the chronic lymphadenopathy and lymphocyte-like cell descriptions. Nevertheless, this study has presented some limitations for NLP, including the failure of NLP to interpret correctly outdated medical terms such as globulins. Altogether, the NLP-driven tools could support the diagnosis but not replace the clinical expertise, especially in case analysis (84).

Another study has applied NLP to distinguish between patients with aggressive haematological malignancy in terms of having limited versus adequate social support by using it to unstructured clinical notes in EHRs. The result showed that patients with limited social support, as identified by NLP, had significantly worse overall survival and a higher risk of hospital readmission or death within 90 days. These findings not only demonstrate the importance of social support in disease prognosis but also highlight the power of NLP in extracting non-clinical variables from health records without directly surveying patients (85).

Other research has investigated the potential of an AI-driven approach to identify clinical trial eligibility criteria for relapsed/refractory MM (RRMM) that can be safely relaxed to enhance trial inclusivity. For this, real-world data from two large datasets and eligibility criteria from nine historical phase III RRMM trials were used. In this study, an algorithm, referred to as AI pathfinder, was applied to estimate the effect of relaxing specific inclusion and exclusion criteria on progression-free survival hazard ratios. The results depicted that some of the criteria could be safely excluded with no significant effect on trial outcomes, including human immunodeficiency virus (HIV), cardiac problems, and asthma, while other criteria were critical and needed to remain strict. In conclusion, AI-driven approaches can facilitate evidence-based adjustments in trial design; however, caution is highly warranted considering the limitations and variability in real-world datasets (86).

Application of AI in accurate treatment strategies and prognosis for patients with haematological malignancies

Biomarker detection

A biomarker is an essential test for patients with haematological malignancies that provides physicians with several treatment options, resistance to specific therapies, and the anticipated progression of the disease or relapse (87).

Several advancements in AI tools are underway to identify more accurate haematological malignancy biomarkers and monitor minimal residual disease at different follow-up intervals (Table 2). For example, one study has found that using an ML tool has improved the prediction of treatment sensitivity in patients with MM based on gene expression data (88). Another study has utilised ML tool as potential targets for subsequent therapy and as biomarkers for prognosis in asymptomatic carriers of adult T-cell leukaemia/lymphoma by characterising a variety of messenger RNA (mRNAs) and microRNAs (miRNAs) and identifying reliable miRNA-mRNA interactions for each subtype (89).

Guerrero et al. [2022] have optimised ML model to predict undetectable MRD at the onset of MM patients. It aims to integrate cytogenetic abnormalities [t(4; 14) and/or del (17p13)], the clonal size of BM plasma cells, circulating tumour plasma cells, and immune biomarkers to develop functional, integrated weighted models for detecting MRD in MM (90).

ML-based analysis of patient characteristics enables more objective risk assessments, supporting the development of targeted therapies tailored to specific biomarkers and promoting personalised patient care (107).

Drug discovery

AI tools have transformed the field of drug discovery by accelerating progress through automation, boosting predictive capabilities, and enhancing efficiency (108) (Table 2).

One study aimed to construct an ML tool to identify signature gene expression specific to leukaemia stem cells and found several genes, such as RFC4 and RFC, that exhibit strong cluster interactions and serve as inhibitory therapeutic targets for AML patients (91). Another study utilised the t-distributed stochastic neighbour embedding (t-SNE) AI algorithm as a mapping model to develop a drug discovery platform capable of detecting the activity of novel kinase inhibitors within the kinome, such as the new inhibitor of FLT3 (25). However, another study built a drug-discovery platform based on an ANN AI algorithm to demonstrate the critical role of the commercial drug simvastatin in targeting the BM microenvironment in treating CLL cell lines. They found that simvastatin was the most effective drug in reducing the survival, proliferation, and adhesion of CLL cells, as well as enhancing the effect of other antitumor agents, such as venetoclax and ibrutinib (92).

Overall, previous research has demonstrated AI’s ability to develop platforms that support the creation of new medicines or combination therapies for blood diseases. This significantly improves treatment outcomes and patients’ prognosis (109).

Disease prognosis and risk assessment

Aggressive leukaemia, such as AML, ALL, and MDS, tends to advance rapidly and is often associated with unfavourable outcomes. However, research has shown that certain abnormalities linked to these conditions may be identifiable before a formal diagnosis (95,110,111). By evaluating a patient’s genetic profile alongside lifestyle habits, AI can efficiently and accurately assess the likelihood of developing specific blood disorders, enabling early medical intervention (Table 2) (112). A recent study has reported the potential of ML algorithms in predicting a high risk of ALL in paediatric patients by considering clinical parameters, patients’ phenotype, and environmental factors (93). Another study has demonstrated the power of ML algorithms in creating an AI personalised prediction model to predict survival and leukaemia transformation possibilities at different time points of MDS patients based on advanced analytics of clinical, pathologic, and molecular data (94).

Several ML algorithms, such as SVMs, ANN, RFs, decision trees, and KNN, can be used to develop different prognosis systems to optimise clinical treatments (113). The recent studies have utilised explainable AI (XAI) methods, such as Local Interpretable Model-Agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP), which are designed to help clinicians better understand the impact of risk factors and the prognosis of the diseases (114,115).

Personalised therapy and outcome prediction

Data from molecular biology, genomics, proteomics, metabolomics, and bioinformatics are continually developed and linked to various layers of biological information, leading to the development of more precise and personalised therapies. AI offers great potential in advancing personalised medicine by enhancing the precision and effectiveness of these customised treatment strategies (Table 2) (116).

One study has successfully developed an AI model based on ML algorithms to predict prognosis in AML patients using key prognostic molecular markers from high-throughput sequencing expression profiling data (96). Other studies have also developed AI models to improve the accuracy of predicting complete remission and survival for individuals with AML (97,98). In addition, ML algorithms can also predict treatment efficiency in paediatric patients with ALL who received radiotherapy by stacking integrated classifiers, helping physicians to track the patients’ status (99).

A recent study in 2024 aimed to identify novel prognostic biomarkers and predict response to therapy based on programmed cell death (PCD) pathways. This study has integrated 13 PCD pathways, such as apoptosis, ferroptosis, and necroptosis, into several ML algorithms to predict prognosis and treatment sensitivity in AML patients. Based on single-cell RNA-seq. They identify six novel genes that are associated with worse prognosis in AML patients; these genes are called pan-PCD-related index (PPCDI) (ATP6V0E2, DDIT4, ETS2, DOCK1, CLEC11A, and HGF). In addition, AML patients with high PPCDI were more resistant to conventional chemotherapy (e.g., doxorubicin) but might be more sensitive to other drugs (e.g., dasatinib, methotrexate) (100).

Another study has set out to develop a multi-omics stratification AI model by integrating 151 RNA-seq datasets, 194 DNA methylation datasets, and 200 somatic mutation datasets from the AML cohort in The Cancer Genome Atlas (TCGA) database. They were able to classify AML patients into four molecular subgroups (CS1–CS4) in which each subgroup exhibited survival rate, clinical features, mutation profiles, immune features, and drug sensitivity prediction. For example, the CS4 subgroup included younger patients with favourable outcomes, had frequent WT1, FLT3, and KIT mutations, and had sensitivity to histone deacetylase (HDAC) and B-cell lymphoma-2 (BCL-2) inhibitors, whereas CS3 classified older patients with poor prognosis and frequent RUNX1, DNMT3A, and TP53 mutations, and were potentially responsive to mammalian target of rapamycin (mTOR) inhibitors. This study highlighted that integrating multi-omics data and AI-based clustering algorithms can improve the prediction of disease outcomes and targeted therapeutic options (101).

Interestingly, another study has aimed to evaluate the knowledge bank of combined genomic, cytogenetic, and clinical data in supporting personalised treatment decisions in 1,500 AML patients to develop multistage prognostic models that estimated the accuracy of remission, relapse, and mortality. The results of this study indicate that it can potentially reduce unnecessary hematopoietic transplantation by up to 25% without affecting the survival rate of those patients. This work showed that combining statistical modelling with a knowledge bank can help to make individualised treatment decisions (117).

Although developed and targeted therapies for leukaemia are gradually being implemented, conventional therapies, such as allogeneic HSCT (allo-HSCT), remain applicable (118). Patients receiving this treatment are highly susceptible to toxicity effects, treatment-related infections, graft-versus-host disease (GVHD), graft failure, and relapse. Thus, an AI tool such as ML has been utilised to predict the risk associated with allo-HSCT treatment, with alternative decision trees showing great potential (102,119). GVHD is a high-priority risk factor that needs to be detected earlier in patients receiving allo-HSCT treatment. Thus, a recent study has applied decision tree tools as AI model to predict the survival rate after allo-HSCT and has successfully recognised seven unique phenotypes of patients with chronic GVHD (119).

In addition, another study has developed a dynamic probabilistic algorithm to predict severe acute GVHD after allo-HSCT from clinical data of 584 adult and 45 paediatric patients. They integrate time-dependent clinical parameters to predict acute GVHD very early. This model has achieved high predictive accuracy in adult test sets. This study has shown how real-time clinical data can enable predicting early risk after allo-HSCT (103).

A study of Bayraktar et al. [2023] has applied unsupervised ML methods to address the limitations of the traditional clinical grading system for GVHD after allo-HCT. The study was conducted in multiple centres cohort of 3,019 patients. This method has a more accurate classification of disease severity compared to traditional grading systems such as the Minnesota criteria. The results showed that the model improved predictive accuracy for overall survival and non-relapse mortality, revealing previously unrecognised subgroups within conventional grades that displayed distinct clinical outcomes (120).

Several new prognostic scoring systems have been innovated using AI tools to improve the accuracy of patient stratification and identify the poorest prognosis. For example, two different scoring systems, the Artificial Intelligence Prognostic Scoring System for Myelofibrosis (AIPSS-MF) and response to ruxolitinib after 6 months (RR6), have been developed to improve the identification of myelofibrosis patients with the poorest prognosis following treatment with Ruxolitinib (95,104). Both systems exceed the performance of traditional models in classifying patients accurately, helping in early identification of those with myelofibrosis who have impaired survival after ruxolitinib treatment. This advancement provides a platform for personalised AI-based prognostic models.

Mass spectrometry-based proteomics has led to the discovery of new potential biomarkers related to disease progression, treatment effects, and chemotherapy resistance in MM (121). The proteasome plays a key role in tumour growth by affecting signalling pathways, immune regulation, and drug resistance (122). One study has developed a Ubiquitin Proteasome Pathway Risk Score (UPPRS) model based on nine genes associated with this pathway. This tool helps classify MM patients and, when used alongside the International Staging System, effectively predicts overall survival outcomes (105). Additionally, a prognostic chart was created to explore mechanisms of resistance to proteasome inhibitors, which may identify novel therapeutic targets within the pathway, potentially lowering relapse rates and improving cure rates in MM (105). Moreover, an ML-based decision support system can analyse proteomic signatures to select the most suitable chemotherapeutic agents for MM patients, thereby enhancing the precision of personalised treatment strategies (123).

Applying extreme gradient boosting (XGBoost) models to 862 patients between 2014 to 2022 has achieved high discrimination performance [area under the curve (AUC) ≈0.83], prognostic evaluation of mantle cell lymphoma. Generally, the study has shown how ML can complement traditional prognostic frameworks by integrating clinical, cytogenetic, and genomic data, thereby improving the prediction of risk assessment in mantle cell lymphoma (106).

Strengths, limitations, and current gaps in the application of AI in haematological malignancies

Advantages and limitations of AI in the diagnosis and treatment of patients with haematological malignancies

The combination of AI and human expertise could advance the intelligence, standardisation, and consistency of haematologic diagnosis and treatment. AI algorithms have proven their capability in interpreting large and complex datasets, improving the accuracy, speed, and objectivity of diagnostic processes and supporting decision-making, particularly for less experienced clinicians. Furthermore, AI systems can generate data from diverse sources, uncover new clinical insights, expand human capabilities, and deepen our understanding of haematologic conditions. Communication between technology and clinical practice could enhance research and lead to better therapeutic outcomes.

Despite the previous benefits, several challenges and limitations exist in applying AI to this field.

Standardised clinical care for haematological malignancy remains a critical and complex area of research that requires continuous improvement. Standardised clinical care refers to delivering consistent treatment protocols based on established clinical guidelines and evidence-based best practices. This approach would help in improving the reliability and accuracy of AI tools, in which unifying diagnosis procedures, image techniques, and reporting standards decreases variation that leads to limited AI model performance and generalizability.

Another limitation is the validation of the AI model. For any AI-based tool to be implemented clinically, it must obtain certification as a medical device, which requires rigorous validation and standardisation. Building and optimising an AI model typically involves applying a considerable amount of data, and this data often needs to be validated using another collection of data. Consequently, this confirms that accessibility to the vital data needs to be improved (124).

In addition, data quality, variability, and image heterogeneity are one of the key limitations of building AI model in the haematological field. Differences in image quality can affect data’s coherence, cause an inconsistent identifying pattern, and potentially lead to incorrect classification or confusion (12,125).

For example, in AI-morphology-based diagnostics, digital analysers have been developed to review blood smears and classify cells automatically. Nevertheless, these analysers often differ significantly across manufacturers due to variations in staining techniques, magnification levels, colour calibration, machine formats, and software compatibility, leading to significant obstacles to standardisation and reliability in clinical use (126). In practice, various research labs and clinical teams utilise different AI platforms that differ in workflow design, testing reagents, and variations in parameter settings, often driven by constraints of institutional infrastructure. This variability in data quality and analytical standards across institutions significantly affects data comparability and model reproducibility (127). Furthermore, the relative autonomy of medical institutions constitutes a major obstacle to data exchange and collaboration, limiting the ability to generalise AI models across diverse clinical settings.

As we mentioned earlier, applying uniform analytical procedures or parameters is fundamental to reducing variability and improving data consistency.

Moreover, AI systems rely on continuous retraining and model updating during evolution to enhance performance, which introduces challenges in maintaining consistency in the clinical environment. These variables include batch effects, differences in sample collection across medical centres, and sample preparation protocols, factors that can introduce bias and complicate algorithm comparison and validation (128). Therefore, integrating AI into clinical practice does not eliminate the need for human oversight. On the contrary, it underscores the importance of clinician involvement. It highlights the necessity for clear guidelines and standardised evaluation criteria to ensure the safe and effective deployment of AI technologies in haematology.

Data serves as the cornerstone of AI applications, and the global diversity of the population offers the potential to build expansive and representative datasets. It is well established that increasing the size of data samples significantly improves the accuracy and performance of AI models. However, in the context of blood disorders, training data is often scarce and highly complex, making the process of accumulating large, high-quality datasets both time-consuming and labour-intensive. Additionally, incomplete diagnosis of morphological, genetic, or molecular features due to variations in equipment, protocols, and personnel errors could affect the accuracy and performance of AI models.

For the practical application of ML in diagnosing and treating haematological diseases, comprehensive and high-dimensional datasets are essential for generating accurate and reliable predictions to inform treatment planning (97). Ethical concerns related to privacy and data security when applying ML techniques to sensitive and confidential patient data may arise due to the lack of standardised procedures (129). Therefore, it is essential to construct high-quality datasets in compliance with relevant regulations, such as the European Union’s General Data Protection Regulation (GDPR) (130). Strengthening the integration between laboratory information systems (LIS) and EHRs, building a secure and mature data-sharing infrastructure, and developing a unified data standardisation framework are all crucial steps toward ensuring data protection and effective sharing. With such a foundation, ML can be more efficiently used to support clinical decision-making, expedite drug discovery, and reduce failure rates in the treatment of haematological malignancies (131).

Another significant challenge is the limited connection between the fields of medicine and engineering. The needs of ongoing AI model development present challenges for clinicians and researchers to be engaged effectively in model design and validation (132). A common methodological issue arises during model development when datasets are categorised into training and test sample subsets. For example, if samples from the same patient appear in both sets, that could lead to overestimating the model’s performance due to data leakage. Additionally, unrecognised variables can produce confounding effects, further reducing generalizability (133). To overcome this limitation and ensure reliable evaluation, the appropriate data-splitting strategies, such as stratified sampling, cross-validation, and external validation, should be applied. Furthermore, employing multiple ML models for comparison can aid in identifying consistent predictive features and verifying the robustness of results across different algorithmic frameworks, rather than serving as a correction for partitioning errors (134).

AI model outputs can also be biased, and developers may misinterpret the intended clinical objectives. Thus, the expertise of seasoned professionals and adherence to standardised diagnostic and treatment protocols are crucial for ensuring the practical application of ML in hematologic care. However, there is currently a shortage of experienced haematologists, and few medical professionals possess the necessary knowledge and technical skills to contribute to AI research and development. Training these interdisciplinary, high-level specialists requires a significant amount of time and financial investment. This talent gap poses a substantial barrier to enhancing the accuracy, utility, and clinical adoption of AI technologies in haematology.

The “black box” characteristic of AI technology is a widely recognised limitation, primarily referring to the inherent lack of interpretability in how complex models reach their decisions. AI systems rely on continuous improvement of ML performance, which generally involves three main components: algorithms, training data, and resulting models. Algorithms are sets of rules or procedures that, when trained on large datasets, learn to identify patterns and perform different tasks. DL models, in particular, comprise a considerable number of parameters that enable them to process high-dimensional data and extract feature representations. However, this complexity often renders the model’s internal processes opaque, making it difficult to understand the true essence of the “black box” problem (135).

It is essential to mention that, in addition to this intrinsic opacity, external factors may further limit transparency. Developers may deliberately restrict access to training data, algorithms, or model details to protect intellectual property or enhance performance. While these practices contribute to the perception of AI systems as “black boxes”, they are conceptually distinct from the fundamental interpretability challenge posed by the models themselves. Together, these factors make it difficult for non-experts, such as clinicians, to understand or trust AI-driven decisions, raising concerns about reliability, safety, and ultimately hindering broader clinical adoption.

To overcome the limitations, recent research has focused on XAI technologies that are designed to capture and clarify the outputs of ML/DL algorithms, thereby rendering the decision-making process more transparent and interpretable (114,115,136). These technologies have demonstrated a significant role in supporting diagnosis, drug discovery, and development (137,138). However, XAI remains in development and lacks comprehensive standardisation.

Limitations and gaps in the current literature

Several drawbacks in the literature on the role of AI in the diagnosis, prognosis, and treatment of haematological malignancies have been noted during the preparation of this review. One of the major concerns is data bias, as many AI models are trained on datasets that lack diversity in patient demographics or disease subtypes, limiting their generalizability across populations. Additionally, most study designs are retrospective, increasing the risk of selection bias and reducing the reliability of causal conclusions. Another concern is a lack of external validation, with many models showing high accuracy only within the original dataset, raising concerns about overfitting and poor performance in real-world clinical settings. The heterogeneity in AI methodologies, including differences in algorithms, feature selection, and outcome measures, makes it difficult to compare studies or establish best practices. In addition, most studies do not evaluate the actual clinical impact or integration of AI tools, and few address regulatory, ethical, or data privacy issues, which are essential for safe and effective implementation in haematology. Finally, almost all the AI models included in this review lack global regulatory approval for medical use. These limitations underscore the need for more rigorous, standardised, and ethically guided research to support the reliable use of AI in haematologic oncology.

Conclusions

ML and DL algorithms are increasingly applied to AI models for the comprehensive diagnosis and precision treatment of haematologic malignancies. These technologies offer robust data processing capabilities that can significantly accelerate laboratory diagnostics, often achieving accuracy levels comparable to or even exceeding those of human experts.

Managing complications of blood disorders, such as drug resistance, GVHD following allo-HSCT, and disease relapse, frequently presents challenges. By integrating diverse clinical datasets with ML techniques, AI has assisted in uncovering underlying disease mechanisms, resistance pathways, and novel therapeutic targets. This integration facilitates more precise drug development, risk stratification, and continuous monitoring of patient prognosis.

Despite all previous challenges of AI technology, significant progress has been made. These advancements encompass the creation of digital databases for haematological conditions, the establishment of ethical and privacy regulations for AI usage, the inclusion of AI education in medical school curricula, and the development of more interpretable AI models.

In essence, AI holds great promise for improving the accuracy and effectiveness of clinical decision-making in haematology. However, the aim is not to replace healthcare professionals but to enhance and standardise diagnostic and treatment processes through AI-assisted tools. Looking ahead, further research is essential to develop a fully integrated and intelligent platform for diagnosis and therapy that aligns seamlessly with established clinical pathways.

Acknowledgments

The author would like to acknowledge the support of the Department of Laboratory Medicine, Faculty of Applied Medical Sciences, Al-Baha University, for providing the academic environment that facilitated the completion of this review.

Footnote

Reporting Checklist: The author has completed the Narrative Review reporting checklist. Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-195/rc

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-195/prf

Funding: None.

Conflicts of Interest: The author has completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2025-195/coif). The author has no conflicts of interest to declare.

Ethical Statement: The author is accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Haque R, Al Sakib A, Hossain MF, et al. Advancing early leukemia diagnostics: A comprehensive study incorporating image processing and transfer learning. BioMedInformatics 2024;4:966-91.
2. Auberger P, Tamburini-Bonnefoy J, Puissant A. Drug Resistance in Hematological Malignancies. Int J Mol Sci 2020;21:6091. [Crossref] [PubMed]
3. Gkirkas K, Stamouli M, Karagiannidou A, et al. Extramedullary leukemia relapse after allogeneic stem cell transplantation: a novel mechanism of immune escape? Immunotherapy 2020;12:635-40. [Crossref] [PubMed]
4. Zini G. Artificial intelligence in hematology. Hematology 2005;10:393-400. [Crossref] [PubMed]
5. Samuel AL. Some studies in machine learning using the game of checkers. II—recent progress. Annual Review in Automatic Programming 1969;6:1-36.
6. Liu K, Hu J. Classification of acute myeloid leukemia M1 and M2 subtypes using machine learning. Comput Biol Med 2022;147:105741. [Crossref] [PubMed]
7. El Achi H, Khoury JD. Artificial Intelligence and Digital Microscopy Applications in Diagnostic Hematopathology. Cancers (Basel) 2020;12:797. [Crossref] [PubMed]
8. Mienye ID, Swart TG. A comprehensive review of deep learning: Architectures, recent advances, and applications. Information 2024;15:755.
9. Purwar S, Tripathi R, Ranjan R, et al. Classification of thalassemia patients using a fusion of deep image and clinical features. In: 2021 11th International Conference on Cloud Computing, Data Science & Engineering (Confluence). IEEE; 2021:410-5.
10. Vogado LHS, Veras RMS, Araujo FHD, et al. Leukemia diagnosis in blood slides using transfer learning in CNNs and SVM for classification. Eng Appl Artif Intell 2018;72:415-22.
11. Gedefaw L, Liu CF, Ip RKL, et al. Artificial Intelligence-Assisted Diagnostic Cytology and Genomic Testing for Hematologic Disorders. Cells 2023;12:1755. [Crossref] [PubMed]
12. Dese K, Raj H, Ayana G, et al. Accurate Machine-Learning-Based classification of Leukemia from Blood Smear Images. Clin Lymphoma Myeloma Leuk 2021;21:e903-14. [Crossref] [PubMed]
13. Jia Y, Li C, Feng C, et al. Prognostic prediction for inflammatory breast cancer patients using random survival forest modeling. Transl Oncol 2025;52:102246. [Crossref] [PubMed]
14. Handelman GS, Kok HK, Chandra RV, et al. Peering Into the Black Box of Artificial Intelligence: Evaluation Metrics of Machine Learning Methods. AJR Am J Roentgenol 2019;212:38-43. [Crossref] [PubMed]
15. Lin E, Fuda F, Luu HS, et al. Digital pathology and artificial intelligence as the next chapter in diagnostic hematopathology. Semin Diagn Pathol 2023;40:88-94. [Crossref] [PubMed]
16. Katz BZ, Feldman MD, Tessema M, et al. Evaluation of Scopio Labs X100 Full Field PBS: The first high-resolution full field viewing of peripheral blood specimens combined with artificial intelligence-based morphological analysis. Int J Lab Hematol 2021;43:1408-16. [Crossref] [PubMed]
17. Makhija K, Lincz LF, Attalla K, et al. White blood cell evaluation in haematological malignancies using a web-based digital microscopy platform. Int J Lab Hematol 2021;43:1379-87. [Crossref] [PubMed]
18. Fu X, Fu M, Li Q, et al. Morphogo: An Automatic Bone Marrow Cell Classification System on Digital Images Analyzed by Artificial Intelligence. Acta Cytol 2020;64:588-96. [Crossref] [PubMed]
19. Zhong P, Hong M, He H, et al. Diagnosis of Acute Leukemia by Multiparameter Flow Cytometry with the Assistance of Artificial Intelligence. Diagnostics (Basel) 2022;12:827. [Crossref] [PubMed]
20. Qin Y, Wen J, Zheng H, et al. Varifocal-Net: A Chromosome Classification Approach Using Deep Convolutional Networks. IEEE Trans Med Imaging 2019;38:2569-81. [Crossref] [PubMed]
21. Xia C, Wang J, Qin Y, et al. KaryoNet: Chromosome Recognition With End-to-End Combinatorial Optimization Network. IEEE Trans Med Imaging 2023;42:2899-911. [Crossref] [PubMed]
22. Cao F, Zhang Y, Cai Y, et al. Chromatin interaction neural network (ChINN): a machine learning-based method for predicting chromatin interactions from DNA sequences. Genome Biol 2021;22:226. [Crossref] [PubMed]
23. Chen ZH, Lin L, Wu CF, et al. Artificial intelligence for assisting cancer diagnosis and treatment in the era of precision medicine. Cancer Commun (Lond) 2021;41:1100-15. [Crossref] [PubMed]
24. Zhang Z, Wei X. Artificial intelligence-assisted selection and efficacy prediction of antineoplastic strategies for precision cancer therapy. Semin Cancer Biol 2023;90:57-72. [Crossref] [PubMed]
25. Janssen APA, Grimm SH, Wijdeven RHM, et al. Drug Discovery Maps, a Machine Learning Model That Visualizes and Predicts Kinome-Inhibitor Interaction Landscapes. J Chem Inf Model 2019;59:1221-9. [Crossref] [PubMed]
26. Walter W, Pohlkamp C, Meggendorfer M, et al. Artificial intelligence in hematological diagnostics: Game changer or gadget? Blood Rev 2023;58:101019. [Crossref] [PubMed]
27. Riley RS, Hogan TF, Pavot DR, et al. A pathologist's perspective on bone marrow aspiration and biopsy: I. Performing a bone marrow examination. J Clin Lab Anal 2004;18:70-90. [Crossref] [PubMed]
28. Adewoyin AS, Nwogoh B. Peripheral blood film - a review. Ann Ib Postgrad Med 2014;12:71-9.
29. Khoury JD, Solary E, Abla O, et al. The 5th edition of the World Health Organization Classification of Haematolymphoid Tumours: Myeloid and Histiocytic/Dendritic Neoplasms. Leukemia 2022;36:1703-19.
30. Kim H, Hur M, d'Onofrio G, et al. Real-World Application of Digital Morphology Analyzers: Practical Issues and Challenges in Clinical Laboratories. Diagnostics (Basel) 2025;15:677. [Crossref] [PubMed]
31. Dasariraju S, Huo M, McCalla S. Detection and Classification of Immature Leukocytes for Diagnosis of Acute Myeloid Leukemia Using Random Forest Algorithm. Bioengineering (Basel) 2020;7:120. [Crossref] [PubMed]
32. Sazak H, Kotan M. Automated Blood Cell Detection and Classification in Microscopic Images Using YOLOv11 and Optimized Weights. Diagnostics (Basel) 2024;15:22. [Crossref] [PubMed]
33. Wang M, Dong C, Gao Y, et al. A Deep Learning Model for the Automatic Recognition of Aplastic Anemia, Myelodysplastic Syndromes, and Acute Myeloid Leukemia Based on Bone Marrow Smear. Front Oncol 2022;12:844978. [Crossref] [PubMed]
34. Arbab Q, Khan MQ, Ali H. Automatic Detection and Classification of Acute Lymphoblastic Leukemia Using Convolution Neural Network. The Sciencetech 2022;3:60-72.
35. Wei BH, Tsai XC, Sun KJ, et al. Annotation-free deep learning for predicting gene mutations from whole slide images of acute myeloid leukemia. NPJ Precis Oncol 2025;9:35. [Crossref] [PubMed]
36. Cheng ZJ, Li H, Liu M, et al. Artificial intelligence reveals the predictions of hematological indexes in children with acute leukemia. BMC Cancer 2024;24:993. [Crossref] [PubMed]
37. Khosla E, Ramesh D. Phase classification of chronic myeloid leukemia using convolution neural networks. In: 2018 4th International Conference on Recent Advances in Information Technology (RAIT). IEEE; 2018:1-6.
38. Eckardt JN, Middeke JM, Riechert S, et al. Deep learning detects acute myeloid leukemia and predicts NPM1 mutation status from bone marrow smears. Leukemia 2022;36:111-8. [Crossref] [PubMed]
39. Kockwelp J, Thiele S, Bartsch J, et al. Deep learning predicts therapy-relevant genetics in acute myeloid leukemia from Pappenheim-stained bone marrow smears. Blood Adv 2024;8:70-9. [Crossref] [PubMed]
40. Wang YF, Li GH, Hanson JR, et al. Machine learning-facilitated cross-panel automatic classification at both sample and cell level in acute myeloid leukemia. Blood 2023;142:5019.
41. Lu Z, Morita M, Yeager TS, et al. Validation of Artificial Intelligence (AI)-Assisted Flow Cytometry Analysis for Immunological Disorders. Diagnostics (Basel) 2024;14:420. [Crossref] [PubMed]
42. Cheng FM, Lo SC, Lin CC, et al. Deep learning assists in acute leukemia detection and cell classification via flow cytometry using the acute leukemia orientation tube. Sci Rep 2024;14:8350. [Crossref] [PubMed]
43. Ni W, Tong X, Qian W, et al. Discrimination of malignant neutrophils of chronic myelogenous leukemia from normal neutrophils by support vector machine. Comput Biol Med 2013;43:1192-5. [Crossref] [PubMed]
44. McCarthy N, Gui G, Dumezy F, et al. Pre-emptive detection and evolution of relapse in acute myeloid leukemia by flow cytometric measurable residual disease surveillance. Leukemia 2024;38:1667-73. [Crossref] [PubMed]
45. Bokhari Y, Alhareeri A, Aljouie A, et al. ChromoEnhancer: An Artificial-Intelligence-Based Tool to Enhance Neoplastic Karyograms as an Aid for Effective Analysis. Cells 2022;11:2244. [Crossref] [PubMed]
46. Liu X, Fu L, Chun-Wei Lin J, et al. SRAS-net: Low-resolution chromosome image classification based on deep learning. IET Syst Biol 2022;16:85-97. [Crossref] [PubMed]
47. Warnat-Herresthal S, Perrakis K, Taschler B, et al. Scalable Prediction of Acute Myeloid Leukemia Using High-Dimensional Machine Learning and Blood Transcriptomics. iScience 2020;23:100780. [Crossref] [PubMed]
48. Vasighizaker A, Sharma A, Dehzangi A. A novel one-class classification approach to accurately predict disease-gene association in acute myeloid leukemia cancer. PLoS One 2019;14:e0226115. [Crossref] [PubMed]
49. Zhong FM, Yao FY, Yang YL, et al. Molecular subtypes predict therapeutic responses and identifying and validating diagnostic signatures based on machine learning in chronic myeloid leukemia. Cancer Cell Int 2023;23:61. [Crossref] [PubMed]
50. Levy JJ, Chen Y, Azizgolshani N, et al. MethylSPWNet and MethylCapsNet: Biologically Motivated Organization of DNAm Neural Networks, Inspired by Capsule Networks. NPJ Syst Biol Appl 2021;7:33. [Crossref] [PubMed]
51. Levy JJ, Titus AJ, Petersen CL, et al. MethylNet: an automated and modular deep learning approach for DNA methylation analysis. BMC Bioinformatics 2020;21:108. [Crossref] [PubMed]
52. Tojo A. Clinical sequencing in leukemia with the assistance of artificial intelligence. Rinsho Ketsueki 2017;58:1913-7. [Crossref] [PubMed]
53. Yokoyama K. Artificial intelligence-guided precision medicine in hematological disorders. Rinsho Ketsueki 2020;61:554-63. [Crossref] [PubMed]
54. Schmidt F, Ranjan B, Lin QXX, et al. RCA2: a scalable supervised clustering algorithm that reduces batch effects in scRNA-seq data. Nucleic Acids Res 2021;49:8505-19. [Crossref] [PubMed]
55. Wang T, Johnson TS, Shao W, et al. BERMUDA: a novel deep transfer learning method for single-cell RNA sequencing batch correction reveals hidden high-resolution cellular subtypes. Genome Biol 2019;20:165. [Crossref] [PubMed]
56. Villiers W, Kelly A, He X, et al. Multi-omics and machine learning reveal context-specific gene regulatory activities of PML::RARA in acute promyelocytic leukemia. Nat Commun 2023;14:724. [Crossref] [PubMed]
57. Liang CA, Chen L, Wahed A, et al. Proteomics Analysis of FLT3-ITD Mutation in Acute Myeloid Leukemia Using Deep Learning Neural Network. Ann Clin Lab Sci 2019;49:119-26.
58. Yu TC, Yang CK, Hsu WH, et al. A machine-learning-based algorithm for bone marrow cell differential counting. Int J Med Inform 2025;194:105692. [Crossref] [PubMed]
59. Hameed M, Raja MAZ, Zameer A, et al. Acute myeloid leukemia classification using ReLViT and detection with YOLO enhanced by adversarial networks on bone marrow images. Sci Rep 2025;15:32798. [Crossref] [PubMed]
60. Anand V, Bachhal P, Koundal D, et al. Deep learning model for early acute lymphoblastic leukemia detection using microscopic images. Sci Rep 2025;15:29147. [Crossref] [PubMed]
61. Katz BZ, Benisty D, Sayegh Y, et al. Remote Digital Microscopy Improves Hematology Laboratory Workflow by Reducing Peripheral Blood Smear Analysis Turnaround Time. Appl Clin Inform 2022;13:1108-15. [Crossref] [PubMed]
62. Campana D, Behm FG. Immunophenotyping of leukemia. J Immunol Methods 2000;243:59-75. [Crossref] [PubMed]
63. Porwit A, Béné MC, Duetz C, et al. Multiparameter flow cytometry in the evaluation of myelodysplasia: Analytical issues: Recommendations from the European LeukemiaNet/International Myelodysplastic Syndrome Flow Cytometry Working Group. Cytometry B Clin Cytom 2023;104:27-50. [Crossref] [PubMed]
64. Lewis JE, Cooper LAD, Jaye DL, et al. Automated Deep Learning-Based Diagnosis and Molecular Characterization of Acute Myeloid Leukemia Using Flow Cytometry. Mod Pathol 2024;37:100373. [Crossref] [PubMed]
65. Ujfalusi A. The role of cytogenetic tests in the diagnosis of malignant hematologic diseases. Magy Onkol 2024;68:341.
66. Bayani J, Squire JA. Traditional banding of chromosomes for cytogenetic analysis. Curr Protoc Cell Biol 2004;Chapter 22:Unit 22.3.
67. Walter W, Haferlach C, Nadarajah N, et al. How artificial intelligence might disrupt diagnostics in hematology in the near future. Oncogene 2021;40:4271-80. [Crossref] [PubMed]
68. Nimitha N, Ezhumalai P, Chokkalingam A. An improved deep convolutional neural network architecture for chromosome abnormality detection using hybrid optimization model. Microsc Res Tech 2022;85:3115-29. [Crossref] [PubMed]
69. Vajen B, Hänselmann S, Lutterloh F, et al. Classification of fluorescent R-Band metaphase chromosomes using a convolutional neural network is precise and fast in generating karyograms of hematologic neoplastic cells. Cancer Genet 2022;260-261:23-9. [Crossref] [PubMed]
70. Dwivedi S, Purohit P, Misra R, et al. Diseases and Molecular Diagnostics: A Step Closer to Precision Medicine. Indian J Clin Biochem 2017;32:374-98. [Crossref] [PubMed]
71. Koutsi A, Vervesou EC. Diagnostic molecular techniques in haematology: recent advances. Ann Transl Med 2018;6:242. [Crossref] [PubMed]
72. Reilly B, Tanaka TN, Diep D, et al. DNA methylation identifies genetically and prognostically distinct subtypes of myelodysplastic syndromes. Blood Adv 2019;3:2845-58. [Crossref] [PubMed]
73. Koldobskiy MA, Jenkinson G, Abante J, et al. Converging genetic and epigenetic drivers of paediatric acute lymphoblastic leukaemia identified by an information-theoretic analysis. Nat Biomed Eng 2021;5:360-76. [Crossref] [PubMed]
74. Teschendorff AE, Zhu T, Breeze CE, et al. EPISCORE: cell type deconvolution of bulk tissue DNA methylomes from single-cell RNA-Seq data. Genome Biol 2020;21:221. [Crossref] [PubMed]
75. Zhu T, Teschendorff AE. Cell-Type Deconvolution of Bulk DNA Methylation Data with EpiSCORE. Methods Mol Biol 2023;2629:23-42. [Crossref] [PubMed]
76. Lupiáñez DG, Kraft K, Heinrich V, et al. Disruptions of topological chromatin domains cause pathogenic rewiring of gene-enhancer interactions. Cell 2015;161:1012-25. [Crossref] [PubMed]
77. Xu C, Jackson SA. Machine learning and complex biological data. Genome Biol 2019;20:76. [Crossref] [PubMed]
78. Rawoof A, Swaminathan G, Tiwari S, et al. LeukmiR: a database for miRNAs and their targets in acute lymphoblastic leukemia. Database (Oxford) 2020;2020:baz151. [Crossref] [PubMed]
79. Hong M, Tao S, Zhang L, et al. RNA sequencing: new technologies and applications in cancer research. J Hematol Oncol 2020;13:166. [Crossref] [PubMed]
80. Lei Y, Tang R, Xu J, et al. Applications of single-cell sequencing in cancer research: progress and perspectives. J Hematol Oncol 2021;14:91. [Crossref] [PubMed]
81. Sampson DL, Parker TJ, Upton Z, et al. A comparison of methods for classifying clinical samples based on proteomics data: a case study for statistical and machine learning approaches. PLoS One 2011;6:e24973. [Crossref] [PubMed]
82. Vizza P, Aracri F, Guzzi PH, et al. Machine learning pipeline to analyze clinical and proteomics data: experiences on a prostate cancer case. BMC Med Inform Decis Mak 2024;24:93. [Crossref] [PubMed]
83. Au Yeung J, Shek A, Searle T, et al. Natural language processing data services for healthcare providers. BMC Med Inform Decis Mak 2024;24:356. [Crossref] [PubMed]
84. Fan BE, Chow M. Artificial intelligence assisted diagnosis of chronic lymphocytic leukemia in a historic pediatric case using large language models. Ann Hematol 2024;103:3815-9. [Crossref] [PubMed]
85. Johnson PC, Markovitz NH, Gray TF, et al. Association of Social Support With Overall Survival and Healthcare Utilization in Patients With Aggressive Hematologic Malignancies. J Natl Compr Canc Netw 2021;2021:1-7. [Crossref] [PubMed]
86. Jreich R, Zhang H, Meng Z, et al. Evaluating the robustness of an AI pathfinder application on eligibility criteria in multiple myeloma trials using real-world data and historical trials. J Comp Eff Res 2024;13:e230164. [Crossref] [PubMed]
87. Das S, Dey MK, Devireddy R, et al. Biomarkers in Cancer Detection, Diagnosis, and Prognosis. Sensors (Basel) 2023;24:37. [Crossref] [PubMed]
88. Venezian Povoa L, Ribeiro CHC, Silva ITD. Machine learning predicts treatment sensitivity in multiple myeloma based on molecular and clinical information coupled with drug response. PLoS One 2021;16:e0254596. [Crossref] [PubMed]
89. Ghobadi MZ, Emamzadeh R, Afsaneh E. Exploration of mRNAs and miRNA classifiers for various ATLL cancer subtypes using machine learning. BMC Cancer 2022;22:433. [Crossref] [PubMed]
90. Guerrero C, Puig N, Cedena MT, et al. A Machine Learning Model Based on Tumor and Immune Biomarkers to Predict Undetectable MRD and Survival Outcomes in Multiple Myeloma. Clin Cancer Res 2022;28:2598-609. [Crossref] [PubMed]
91. Zhang Y, Liu D, Li F, et al. Identification of biomarkers for acute leukemia via machine learning-based stemness index. Gene 2021;804:145903. [Crossref] [PubMed]
92. Gimenez N, Tripathi R, Giró A, et al. Systems biology drug screening identifies statins as enhancers of current therapies in chronic lymphocytic leukemia. Sci Rep 2020;10:22153. [Crossref] [PubMed]
93. Mahmood N, Shahid S, Bakhshi T, et al. Identification of significant risks in pediatric acute lymphoblastic leukemia (ALL) through machine learning (ML) approach. Med Biol Eng Comput 2020;58:2631-40. [Crossref] [PubMed]
94. Nazha A, Komrokji R, Meggendorfer M, et al. Personalized Prediction Model to Risk Stratify Patients With Myelodysplastic Syndromes. J Clin Oncol 2021;39:3737-46. [Crossref] [PubMed]
95. Duminuco A, Mosquera-Orgueira A, Nardo A, et al. AIPSS-MF machine learning prognostic score validation in a cohort of myelofibrosis patients treated with ruxolitinib. Cancer Rep (Hoboken) 2023;6:e1881. [Crossref] [PubMed]
96. Niu P, Yao B, Wei L, et al. Construction of prognostic risk prediction model based on high-throughput sequencing expression profile data in childhood acute myeloid leukemia. Blood Cells Mol Dis 2019;77:43-50. [Crossref] [PubMed]
97. Eckardt JN, Röllig C, Metzeler K, et al. Prediction of complete remission and survival in acute myeloid leukemia using supervised machine learning. Haematologica 2023;108:690-704. [Crossref] [PubMed]
98. Wagner S, Vadakekolathu J, Tasian SK, et al. A parsimonious 3-gene signature predicts clinical outcomes in an acute myeloid leukemia multicohort study. Blood Adv 2019;3:1330-46. [Crossref] [PubMed]
99. Kashef A, Khatibi T, Mehrvar A. Prediction of Cranial Radiotherapy Treatment in Pediatric Acute Lymphoblastic Leukemia Patients Using Machine Learning: A Case Study at MAHAK Hospital. Asian Pac J Cancer Prev 2020;21:3211-9. [Crossref] [PubMed]
100. Qin Y, Pu X, Hu D, et al. Machine learning-based biomarker screening for acute myeloid leukemia prognosis and therapy from diverse cell-death patterns. Sci Rep 2024;14:17874. [Crossref] [PubMed]
101. Wang T, Cui S, Lyu C, et al. Molecular precision medicine: Multi-omics-based stratification model for acute myeloid leukemia. Heliyon 2024;10:e36155. [Crossref] [PubMed]
102. Fuse K, Uemura S, Tamura S, et al. Patient-based prediction algorithm of relapse after allo-HSCT for acute Leukemia and its usefulness in the decision-making process using a machine learning approach. Cancer Med 2019;8:5058-67. [Crossref] [PubMed]
103. Liu X, Cao Y, Guo Y, et al. Author Correction: Dynamic forecasting of severe acute graft-versus- host disease after transplantation. Nat Comput Sci 2022;2:342. [Crossref] [PubMed]
104. Maffioli M, Mora B, Ball S, et al. A prognostic model to predict survival after 6 months of ruxolitinib in patients with myelofibrosis. Blood Adv 2022;6:1855-64. [Crossref] [PubMed]
105. Ren L, Xu B, Xu J, et al. A Machine Learning Model to Predict Survival and Therapeutic Responses in Multiple Myeloma. Int J Mol Sci 2023;24:6683. [Crossref] [PubMed]
106. Hill HA, Jain P, Ok CY, et al. Integrative Prognostic Machine Learning Models in Mantle Cell Lymphoma. Cancer Res Commun 2023;3:1435-46. [Crossref] [PubMed]
107. Eckardt JN, Bornhäuser M, Wendt K, et al. Application of machine learning in the management of acute myeloid leukemia: current practice and future prospects. Blood Adv 2020;4:6077-85. [Crossref] [PubMed]
108. Kolluri S, Lin J, Liu R, et al. Machine Learning and Artificial Intelligence in Pharmaceutical Research and Development: a Review. AAPS J 2022;24:19. [Crossref] [PubMed]
109. Singh S, Kumar R, Payra S, et al. Artificial Intelligence and Machine Learning in Pharmacological Research: Bridging the Gap Between Data and Drug Discovery. Cureus 2023;15:e44359. [Crossref] [PubMed]
110. Tsai XC, Sun KJ, Lo MY, et al. Poor prognostic implications of myelodysplasia-related mutations in both older and younger patients with de novo AML. Blood Cancer J 2023;13:4. [Crossref] [PubMed]
111. Wachter F, Pikman Y. Pathophysiology of Acute Myeloid Leukemia. Acta Haematol 2024;147:229-46. [Crossref] [PubMed]
112. Gu M, Kovilakam SC, Dunn WG, et al. Multiparameter prediction of myeloid neoplasia risk. Nat Genet 2023;55:1523-30. [Crossref] [PubMed]
113. Pourhomayoun M, Shakibi M. Predicting mortality risk in patients with COVID-19 using machine learning to help medical decision-making. Smart Health (Amst) 2021;20:100178. [Crossref] [PubMed]
114. Yagin FH, Cicek İB, Alkhateeb A, et al. Explainable artificial intelligence model for identifying COVID-19 gene biomarkers. Comput Biol Med 2023;154:106619. [Crossref] [PubMed]
115. Rahimi S, Chu C, Grad R, et al. Explainable Machine Learning Model to Predict COVID-19 Severity Among Older Adults in the Province of Quebec. Ann Fam Med 2023;21:3619. [Crossref] [PubMed]
116. Parekh AE, Shaikh OA. Artificial intelligence (AI) in personalized medicine: AI-generated personalized therapy regimens based on genetic and medical history: short communication. Ann Med Surg (Lond) 2023;85:5831-3. [Crossref] [PubMed]
117. Gerstung M, Papaemmanuil E, Martincorena I, et al. Precision oncology for acute myeloid leukemia using a knowledge bank approach. Nat Genet 2017;49:332-40. [Crossref] [PubMed]
118. Xu ZL, Huang XJ. Optimizing outcomes for haploidentical hematopoietic stem cell transplantation in severe aplastic anemia with intensive GVHD prophylaxis: a review of current findings. Expert Rev Hematol 2021;14:449-55. [Crossref] [PubMed]
119. Gandelman JS, Byrne MT, Mistry AM, et al. Machine learning reveals chronic graft-versus-host disease phenotypes and stratifies survival after stem cell transplant for hematologic malignancies. Haematologica 2019;104:189-96. [Crossref] [PubMed]
120. Bayraktar E, Graf T, Ayuk FA, et al. Data-driven grading of acute graft-versus-host disease. Nat Commun 2023;14:7799. [Crossref] [PubMed]
121. Ho M, Bianchi G, Anderson KC. Proteomics-inspired precision medicine for treating and understanding multiple myeloma. Expert Rev Precis Med Drug Dev 2020;5:67-85. [Crossref] [PubMed]
122. Zhou X, Xu R, Wu Y, et al. The role of proteasomes in tumorigenesis. Genes Dis 2024;11:101070. [Crossref] [PubMed]
123. Katsenou A, O'Farrell R, Dowling P, et al. Using Proteomics Data to Identify Personalized Treatments in Multiple Myeloma: A Machine Learning Approach. Int J Mol Sci 2023;24:15570. [Crossref] [PubMed]
124. Muhsen IN, Shyr D, Sung AD, et al. Machine Learning Applications in the Diagnosis of Benign and Malignant Hematological Diseases. Clin Hematol Int 2021;3:13-20. [Crossref] [PubMed]
125. Ting DSW, Peng L, Varadarajan AV, et al. Deep learning in ophthalmology: The technical and clinical considerations. Prog Retin Eye Res 2019;72:100759. [Crossref] [PubMed]
126. Kratz A, Lee SH, Zini G, et al. Digital morphology analyzers in hematology: ICSH review and recommendations. Int J Lab Hematol 2019;41:437-47. [Crossref] [PubMed]
127. Fan BE, Yong BSJ, Li R, et al. From microscope to micropixels: A rapid review of artificial intelligence for the peripheral blood film. Blood Rev 2024;64:101144. [Crossref] [PubMed]
128. Moser T, Kühberger S, Lazzeri I, et al. Bridging biological cfDNA features and machine learning approaches. Trends Genet 2023;39:285-307. [Crossref] [PubMed]
129. Khalid N, Qayyum A, Bilal M, et al. Privacy-preserving artificial intelligence in healthcare: Techniques and applications. Comput Biol Med 2023;158:106848. [Crossref] [PubMed]
130. Mondschein CF, Monda C. The EU’s General Data Protection Regulation (GDPR) in a Research Context. In: Kubben P, Dumontier M, Dekker A, editors. Fundamentals of Clinical Data Science. Cham: Springer; 2018:chapter 5.
131. Vamathevan J, Clark D, Czodrowski P, et al. Applications of machine learning in drug discovery and development. Nat Rev Drug Discov 2019;18:463-77. [Crossref] [PubMed]
132. Riley P. Three pitfalls to avoid in machine learning. Nature 2019;572:27-9. [Crossref] [PubMed]
133. Kapoor S, Narayanan A. Leakage and the reproducibility crisis in machine-learning-based science. Patterns (N Y) 2023;4:100804. [Crossref] [PubMed]
134. Efthimiou O, Seo M, Chalkou K, et al. Developing clinical prediction models: a step-by-step guide. BMJ 2024;386:e078276. [Crossref] [PubMed]
135. Hassija V, Chamola V, Mahapatra A, et al. Interpreting black-box models: a review on explainable artificial intelligence. Cognitive Computation 2024;16:45-74.
136. Sarkar A, Praetorius JP, Figge MT. Deep learning-based characterization of neutrophil activation phenotypes in ex vivo human Candida blood infections. Comput Struct Biotechnol J 2024;23:1260-73. [Crossref] [PubMed]
137. Zhang Y, Weng Y, Lund J. Applications of Explainable Artificial Intelligence in Diagnosis and Surgery. Diagnostics (Basel) 2022;12:237. [Crossref] [PubMed]
138. Ali S, Akhlaq F, Imran AS, et al. The enlightening role of explainable artificial intelligence in medical & healthcare domains: A systematic literature review. Comput Biol Med 2023;166:107555. [Crossref] [PubMed]