OBJECTIVES: Digital morphology (DM) systems assisted by artificial intelligence are increasingly being introduced into hematology laboratories; however, data on their performance in routine clinical practice for bone marrow aspirates (BMA) remain limited. We evaluated the automated pre-classification generated by a full-field DM system in a large series of BMA samples from patients with onco-hematological disorders. METHODS: We analyzed 350 BMA samples during routine diagnostic activity; they were evaluated using the Siemens Scopio X100 HT system. Automated results were compared with conventional optical microscopy (OM), which served as the reference method for bone marrow differential counts. Results obtained in pre-classification and post-classification were analyzed separately across major cellular lineages and clinically relevant blast thresholds. RESULTS: Sixteen markedly hypercellular samples could not be quantitatively evaluated. In the 334 evaluable BMA samples, the granulocytic series showed very strong correlation between DM and OM at pre-classification (r = 0.85, 95% CI 0.82-0.88), with further improvement after post-classification (r = 0.93, 95% CI 0.91-0.94). Erythroblast percentage showed strong correlation at pre-classification (r = 0.78, 95% CI 0.74-0.82) and very strong correlation after post-classification (r = 0.93, 95% CI 0.92-0.95). Lymphocyte percentage showed moderate correlation at pre-classification (r = 0.55, 95% CI 0.47-0.62) and strong correlation after post-classification (r = 0.78, 95% CI 0.73-0.82). Blast percentage showed strong correlation overall at pre-classification (r = 0.73, 95% CI 0.67-0.77), with very strong correlation in samples with 5% blasts (r = 0.59, 95% CI 0.48-0.68), improving after post-classification (r = 0.84, 95% CI 0.78-0.88). Low-frequency cell populations and challenging hypercellular smears remained more problematic and required expert morphologic review. CONCLUSIONS: In routine BMA evaluation, the Scopio DM system provided good overall performance for the major marrow lineages, with clear improvement after expert post-classification. Its main value lies in supporting a supervised diagnostic workflow rather than replacing expert microscopic assessment, particularly in challenging samples and in low-frequency or morphologically heterogeneous cell populations.
Zini et al. (2026) studied this question.