Résumé L’intelligence artificielle (IA) transforme profondément l’hématologie biologique en exploitant le machine learning et le deep learning pour analyser des données complexes issues de la morphologie, de la cytométrie en flux et de la biologie moléculaire. Les algorithmes permettent aujourd’hui de classer les cellules sanguines, d’interpréter des panels de cytométrie à haut débit et d’intégrer des données génétiques avec des performances proches, voire supérieures, à celles des experts humains. En cytologie, ils favorisent la reconnaissance automatisée des cellules normales ou pathologiques, notamment dans les leucémies aiguës. En cytométrie, l’IA automatise le gating, améliore la détection de populations rares et augmente la reproductibilité. En biologie moléculaire, elle optimise l’interprétation des profils NGS et épigénomiques, soutenant la classification pronostique des hémopathies. Malgré ces avancées, des défis persistent : qualité des données, standardisation, validation clinique et explicabilité. L’avenir réside dans des modèles multimodaux intégrant simultanément données cytologiques, cytométriques, moléculaires et cliniques. Ces innovations redéfinissent le rôle du biologiste médical, désormais acteur central de la supervision, de l’interprétation et de l’éthique de l’IA en hématologie.

Background: Direct oral anticoagulants (DOACs) interfere with many coagulation assays. While in vitro plasma treatment with activated charcoal can overcome such interference, it requires technical support and is unsuitable for high-throughput workflows. Objectives: To circumvent such limitations, we added activated charcoal directly to citrated blood collection tubes and assessed the impact of such an approach on coagulation assays. Methods: DOAC-spiked and patient whole blood samples (with and without DOACs) were treated with activated charcoal. Outcomes were residual DOAC plasma levels, routine coagulation assays, and specialized hemostasis assays (antithrombin, protein C, protein S, lupus anticoagulant testing, and thrombin generation assay). Results: DOACs became undetectable in 118 of 122 samples from patients who received DOAC after 10 minutes of treatment with activated charcoal. DOAC interference with prothrombin time and activated partial thromboplastin time was resolved in almost all samples from patients who received DOAC; fibrinogen, factor V, and D-dimer levels remained unchanged. Whole blood contact with activated charcoal led to a slight prolongation in prothrombin time and activated partial thromboplastin time and a small decrease in fibrinogen level in blood samples from patients without anticoagulated therapies, yet these changes were not clinically relevant. Conversely, the level of endogenous coagulation inhibitors was significantly reduced, and the level of thrombin generation was significantly enhanced. Conclusion: Adding activated charcoal to whole blood corrected DOAC interference with routine coagulation assays with minimal, if any, impact on the results in the absence of DOACs, thus supporting the development of citrated blood collection tubes preloaded with activated charcoal. Their use would accelerate the laboratory workflow and lead to reliable routine hemostasis assay results in all patients receiving DOACs.

Relapse remains the leading cause of mortality in acute myeloid leukemia (AML), largely due to the persistence of therapy-resistant leukemia stem cells (LSCs). However, surface determinants that sustain LSC function and disease aggressiveness remain incompletely defined. Here, we identify the tetraspanin CD81 as a regulator of LSC function, progression and treatment resistance in AML. Analysis of retrospective patient cohorts revealed that high CD81 surface expression is associated with relapse and adverse clinical outcomes in non-core-binding factor AML. Functional studies demonstrated that elevated CD81 expression promotes chemoresistance and enhances leukemic engraftment in immunodeficient mouse models. In vivo gain- and loss-of-function approaches further established that CD81 drives increased leukemia burden and aggressive disease behavior. Notably, CD81 was enriched within LSC-containing subpopulations, where its expression supported LSC maintenance and resistance to chemotherapy. Mechanistically, CD81 promotes chemoresistance and leukemic aggressiveness through pathways linked to LAPTM4B-mediated STAT3 signaling and enhanced adhesion-dependent cellular interactions. These effects were accompanied by increased migration, invasion, and formation of filopodia-like membrane protrusions. Importantly, therapeutic immunotargeting of CD81 significantly reduced leukemic burden while exhibiting a manageable toxicity profile in preclinical models. Collectively, these findings establish CD81 as a clinically relevant surface marker associated with AML relapse and identify CD81-dependent signaling as a therapeutic vulnerability for targeting LSCs and preventing disease recurrence.

Accurate morphological classification of blood leukocytes remains a cornerstone of hematological diagnostics but still relies on manual expertise, leading to variability across laboratories. The CytologIA data challenge was organized to benchmark artificial intelligence (AI) models for automated classification of normal and pathological leukocytes from peripheral blood smears. Twenty hematology laboratories from France, Belgium, and Switzerland contributed to a multicentric, expert-annotated database of 69,168 images encompassing 23 leukocyte classes. A total of 245 teams from academia, hospitals, and industry participated in the challenge. The top-performing model—combining a YOLOX-based detection module with an ensemble of transformer and convolutional classifiers—achieved a balanced accuracy of 0.94 on the hidden test set, markedly outperforming the baseline CNN (0.82). While abundant cell types such as neutrophils were identified with near-perfect accuracy (>0.97), rarer and morphologically overlapping categories remained challenging. All data and models were released openly on data.gouv.fr and GitHub to ensure full reproducibility. CytologIA represents the first large-scale, open, and collaborative AI benchmark in hematological morphology, establishing a new reference for the development of robust, transferable diagnostic algorithms across institutions

Patient characteristics of the national cohort according to the type of PARPi received (olaparib vs. other PARP inhibitors).

Univariate analysis for overall survival from t-MN diagnosis of patients diagnosed with t-MN after OC according to PARPi exposure.

Blood smear and bone marrow aspiration of patient referred for cytopenia post PARPi.

INTRODUCTION The SH2B3 gene encodes the cytoplasmic adaptor protein LNK that acts as a negative regulator of hematopoietic progenitor cell expansion and self-renewal, particularly through its interaction with JAK2 and regulation of the JAK-STAT pathway. In humans, SH2B3 alterations have been identified across a spectrum of hematologic malignancies, especially myeloproliferative neoplasms (MPN), myelodysplastic syndromes (MDS), MDS/MPN overlap syndromes including chronic myelomonocytic leukemia (CMML), and acute lymphoblastic leukemias (ALLs). Despite increasing recognition, their precise clinical relevance and pathogenic role remain incompletely understood. Here, we retrospectively investigated a cohort of 17 833 patients referred for suspicion of a myeloid neoplasm or acute leukemia. Through systematic assessment of SH2B3 as part of routine next-generation sequencing (NGS)-based diagnostic workflows, we aimed to delineate the mutational spectrum and clinical implications of SH2B3 variants. METHODS DNA was extracted from bone marrow or peripheral blood samples and analyzed using a custom targeted NGS panel covering the entire coding region of SH2B3, along with 125 other genes recurrently mutated in hematological malignancies.Clinical and biological data were retrospectively collected from medical records and communication with the referring physicians who ordered the molecular analyses. RESULTS Among the 17 833 individuals tested, 13 423 (75.3%) carried at least one SH2B3 variant. The vast majority (n = 11 589, 86.3%) harbored only the common p.W262R polymorphism (rs3184504).The remaining 1 830 individuals carried 415 unique SH2B3 variants, which were classified hierarchically into four tiers of pathogenicity based on variant type, minor allele frequency in gnomAD v4.1.0 and variant allele frequency in patient samples. This led to the following distribution: 169 individuals with Tier I variants, 86 with Tier II variants, 167 with Tier III variants and 1 451 individuals with Tier IV (other than p.W262R) variants. Overall, 246 unique individuals (~1.4% of the total 17 833) with Tier I and/or Tier II SH2B3 variants were identified, with notable enrichment in CMML (7.2%) and MPN cases (5.6%). Null variants were observed throughout the entire protein, without restriction to specific domains, suggesting they are loss-of-function alterations, likely through protein instability or complete loss of expression. By contrast, missense variants exhibited a non-random distribution, with clear enrichment in the PH and SH2 domains, which are essential for LNK's membrane localization and interaction with JAK2. In the cohort of 246 patients carrying Tier I/II SH2B3 variants, the retained diagnosis was as follow: AML (n=61), MDS (n=56), CMML (n=33), essential thrombocythemia or idiopathic thrombocytosis (n=31), myelofibrosis (n=17), polycythemia vera or idiopathic erythrocytosis (n=17), BCP-ALL (n=4), MDS/MPN with ring sideroblasts and thrombocytosis (n=4), T-cell ALL (n=4), and clonal cytopenia of undetermined significance (n=9). Diagnosis remained undetermined in 10 cases. The overall median age of this cohort was 72 years (IQR 61–78). Among them, 212 SH2B3-mutated patients (86.2%) harbored additional mutations, with a median of 3 co-mutations per patient. Mutations in TET2 (47%), TP53 (10%), and SF3B1 (17%) were associated with significantly lower hemoglobin levels, while RUNX1 (13%) and CBL (9%) mutations were associated with decreased platelet counts. Additionally, 54 patients (22%) carried canonical MPN driver mutations, including 40 with JAK2 V617F, 10 with CALR mutations, and 5 with MPL W515L. These included 15 patientswith AML (25% of SH2B3-mutated AML), suggesting an enrichment of secondary AML transformed from a previous MPN. By contrast, SH2B3 mutation as the sole aberration was associated with higher hemoglobin concentrations (β = 1.95; 95% CI: 0.87–3.02) and platelet counts (β = 0.23 log; 95% CI: 0.05–0.41). Individuals with isolated SH2B3 mutations were often younger and had higher SH2B3 variant allele frequencies. Familial co-segregation was confirmed in three families. CONCLUSION SH2B3 alterations are enriched in CMML and MPNs, including triple-negative cases. These findings underscore the biological and clinical relevance of SH2B3 in myeloid disorders and support its inclusion in diagnostic screening panels, particularly in cases of unexplained thrombocytosis or erythrocytosis.

Older adults with acute myeloid leukemia (AML) have a poor prognosis because frailty and the characteristics of the disease limit the use of intensive chemotherapy (ICT). Treatment with 5-azacitidine (5-AZA) or low-dose cytarabine (Cytarabine) (LDAC) - with or without venetoclax - is currently recommended in this setting. However, we lack real-life data on response rates and treatment outcomes. We conducted a retrospective, multicenter registry study of 279 older adults with AML (median [interquartile range (IQR)] age: 76 [70-81]) having undergone first-line treatment with LDAC (n = 87) or 5-AZA (n = 192) between 2009 and 2019 (i.e. mainly before the venetoclax era) in a university medical center in France. The complete remission rate was 27.3 % overall. After a median follow-up period of 6.9 months, the median [IQR] overall survival (OS) time was shorter in the LDAC group (4.8 months [2.13-14.41]) than in the 5-AZA group (8.9 months [3.2-13.5]; p = 0.046). Ultimately, however, the OS rates were similar in the LDAC and 5-AZA groups (hazard ratio [HR]: 95 % confidence interval [CI]: 1.37 [0.92-2.04], p = 0.12). None of the conventional markers with prognostic value in younger patients receiving ICT (such as those in the European LeukemiaNet classification) appeared to predict the outcome in our population of older patients. Albumin <30 g/L was the only factor that predicted day-30 mortality and OS (adjusted odds ratio [95 %CI]: 6.25 [2.08 - 20.0]; p < 0.001; adjusted HR [95 %CI]: 0.65 [0.44-0.96]; p = 0.030).

Univariate analysis for overall survival from t-MN diagnosis of patients from the national cohort.

Commutation plot visualizing the mutated genes in t-MN after OC according to PARPi treatment.

Allogeneic haematopoietic stem cell transplantation (ASCT) is a curative treatment for acute myeloid leukaemia (AML) but carries a high risk of gonadotoxicity. Ovarian tissue cryopreservation (OTC) offers a fertility preservation option, yet its safety in AML remains uncertain due to the risk of leukaemic cell reintroduction. The FERTILAM pilot study evaluated measurable residual disease (MRD) in ovarian tissue collected at complete remission (CR) from nine AML patients undergoing OTC before ASCT. MRD was assessed using patient-specific clonal markers via droplet digital polymerase chain reaction on DNA and RNA from bone marrow (BM), ovarian cortex and medulla. At CR, MRD-DNA was detected in ovarian cortex of four of nine patients, all with concurrent MRD positivity in BM. Three patients were negative in both BM and ovarian tissue. Paired cortex/medulla analyses showed concordant MRD-DNA results in five of six patients. BM MRD-RNA and MRD-DNA were fully concordant, whereas two discrepancies were observed between MRD-DNA and MRD-RNA in ovarian tissue. These findings suggest potential leukaemic cell persistence in ovarian tissue despite CR and highlight the need for sensitive molecular assays to assess safety prior to ovarian tissue transplantation.

Flow chart of the study displaying the three cohorts.

Melanoma is an aggressive skin tumor whose incidence is rising sharply, and for which the determination of new prognostic factors is a major challenge. In oncology, circulating tumor cells (CTCs) are at the heart of much research, as they represent a source of tumor material obtained non-invasively by liquid biopsy. With this in mind, this prospective, longitudinal study looked at the detection of CTCs in melanoma patients using the flow cytometry technique, and constitutes a proof-of-principle study, as molecular biology is the most widely used technique today to detect CTCs. The labeling strategy showed high sensitivity and specificity for melanoma cells. All 35 patients in the cohort presented at least one CTC at inclusion, demonstrating that the cells circulate regardless of the stage of the disease. However, a significant increase in the number of CTCs was observed in metastatic stages compared with non-metastatic stages. With regard to the main prognostic factors for melanoma, no significant association was found between the number of CTCs and Breslow thickness or the presence of ulceration. This study must be continued in order to increase the size of the sample, with a more consistent longitudinal follow-up, in order to gain a better understanding of the prognostic significance of CTCs.

The evolution of acute myeloid leukemia (AML) classifications has progressively shifted the diagnostic focus toward genetic criteria. Nevertheless, morphology remains a key element in clinical practice, often serving as the initial trigger for additional molecular investigations. The diagnosis of acute erythroleukemia (AEML), initially defined by the FAB group, is no longer recognized as a distinct entity in the latest WHO and ICC classifications. Some studies have indicated that AEML shares similarities with myelodysplastic neoplasms, including a high frequency of TP53 mutations and adverse karyotypes. Here, we conducted a retrospective analysis in adults with AEML defined using historical morphologic criteria (≥ 50% erythroid precursors and ≥ 20% blasts among non-erythroid cells). In contrast to older patients, young adults (18-60 years) exhibit unique genetic profiles including a high prevalence of normal karyotypes (65%), NPM1 (35%) and UBTF (23%) mutations. AEML morphology in NPM1-mutated cases did not impact clinical outcomes but was associated with specific molecular features, including an enrichment of WT1 and cohesin gene mutations. In this age group, our findings support that morphologically defined AEML often corresponds to AML according to current genetic criteria, consistent with recent classification systems that prioritize molecular features over morphology.

TP53 mutations characteristics among t-MN PARPi national cohort.

Waffle chart (%) of patients addressed for cytopenia exploration after OC exposed to a PARPi treatment from cytopenia diagnosis (A) and cytopenia diagnosis based on NGS results (B). Dot plots (C) showing the median WBC, hemoglobin, and platelet counts from patients referred for cytopenia exploration after OC exposed to a PARPi treatment according to t-MN diagnosis (“t-MN”) or not (“Cytopenia”).