cAItomorph: Transformer-Based Hematological Malignancy Prediction from Peripheral Blood Smears in a Real-Word Cohort

Abstract: Peripheral blood smears remain a cornerstone in the diagnosis of hematological neoplasms, offering rapid and valuable insights that inform subsequent diagnostic steps. However, since neoplastic transformations typically arise in the bone marrow, they may not manifest as detectable aberrations in peripheral blood, presenting a diagnostic challenge. In this paper, we introduce cAItomorph, an explainable transformer-based AI model, trained to classify hematological malignancies based on peripheral blood cytomorphology. Our data comprises peripheral blood single-cell images from 6115 patients with diagnoses confirmed by cytomorphology, cytogenetics, molecular genetics, and immunophenotyping from bone marrow samples, and 495 healthy controls, categorized into 22 detailed and 7 coarse classes. cAItomorph leverages the DinoBloom hematology foundation model and aggregates image encodings via a transformer-based architecture into a single vector. It achieves an overall accuracy of 68$\pm$1% (mean$\pm$s.d., 5-fold cross-validation) in 7-disease classification, with F1 scores of 74$\pm$2% for acute leukemia, 75$\pm$3% for myeloproliferative neoplasms and 82$\pm$3% for no malignancy cases. The overall accuracy increases to 84$\pm$1% in top-2 predictions. By analyzing multi-head attentions, we demonstrate clinically relevant cell-level attentions and pixel-level heatmaps. Moreover, our model's calibrated prediction probabilities reduced the false discovery rate from 13.8% to 12% without missing any acute leukemia cases, thereby decreasing the number of unnecessary bone marrow aspirations. Our code, test data, and model weights are publicly available to ensure reproducibility. This study highlights the potential of AI-assisted diagnostics in hematological malignancies, illustrating how models trained on real-world data could enhance diagnostic accuracy and reduce invasive procedures.