Detection of acute promyelocytic leukemia in peripheral blood and bone marrow with annotation-free deep learning

Abstract

While optical microscopy inspection of blood films and bone marrow aspirates by a hematologist is a crucial step in establishing diagnosis of acute leukemia, especially in low-resource settings where other diagnostic modalities are not available, the task remains time-consuming and prone to human inconsistencies. This has an impact especially in cases of Acute Promyelocytic Leukemia (APL) that require urgent treatment. Integration of automated computational hematopathology into clinical workflows can improve the throughput of these services and reduce cognitive human error. However, a major bottleneck in deploying such systems is a lack of sufficient cell morphological object-labels annotations to train deep learning models. We overcome this by leveraging patient diagnostic labels to train weakly-supervised models that detect different types of acute leukemia. We introduce a deep learning approach, Multiple Instance Learning for Leukocyte Identification (MILLIE), able to perform automated reliable analysis of blood films with minimal supervision. Without being trained to classify individual cells, MILLIE differentiates between acute lymphoblastic and myeloblastic leukemia in blood films. More importantly, MILLIE detects APL in blood films (AUC 0.94 ± 0.04) and in bone marrow aspirates (AUC 0.99 ± 0.01). MILLIE is a viable solution to augment the throughput of clinical pathways that require assessment of blood film microscopy.

Schistocyte detection in artificial intelligence age

“…A group of machine learning approaches have yielded achievements in the identification of myeloid blasts, lymphoblasts, and promyelocytes in acute leukemia, as well as the characterization of morphological variants of PB elements (e.g., features of dysplasia). [18][19][20][21][22] Kimura et al 22 developed an automated diagnostic support system for myelodysplastic syndromes (MDS) based on CNNs and extreme gradient boosting (XGBoost), to differentiate MDS from aplastic anemia (AA) with high accuracy.…”

“…Kassim et al 17 proposed a novel pipeline named RBCNet for erythrocyte and malaria detection in thin blood smear microscopy images, using a dual deep learning architecture, and achieved an accuracy higher than 97%, with a notably higher true positive and lower false alarm rates. A group of machine learning approaches have yielded achievements in the identification of myeloid blasts, lymphoblasts, and promyelocytes in acute leukemia, as well as the characterization of morphological variants of PB elements (e.g., features of dysplasia) 18–22 . Kimura et al 22 developed an automated diagnostic support system for myelodysplastic syndromes (MDS) based on CNNs and extreme gradient boosting (XGBoost), to differentiate MDS from aplastic anemia (AA) with high accuracy.…”

Applied machine learning in hematopathology

“…35 In order to overcome the lack of sufficient labeled cell images of different morphologies, Manescu et al trained a weakly supervised model to classify subtypes of acute leukemia. 36 AI/ML have also been employed to classify acute lymphoblastic leukemia (ALL) subtypes, 37 and multiple myeloma detection. 38 Convolutional neural networks were employed to extract morphological characteristics from 334 MDS, MDS/MPN, and control BM biopsies, which were then utilized to predict genetic, cytogenetic anomalies, as well as prognosis using multivariate regression models.…”

“…Unfortunately, there is a lack of publicly available expert annotated data in hematopathology, which is a problem affecting all pathology subspecialties 35 . In order to overcome the lack of sufficient labeled cell images of different morphologies, Manescu et al trained a weakly supervised model to classify subtypes of acute leukemia 36 . AI/ML have also been employed to classify acute lymphoblastic leukemia (ALL) subtypes, 37 and multiple myeloma detection 38 .…”

Comparison of Deep Learning Techniques in Detection of Sickle Cell Disease

“…We used secondary dataset obtained from an online dataset for digitized thin blood films of sickle cell disease detection, these images were collected by (Manescu et al, 2023). These images were collected using a custom built brightfield microscope fitted with a 100X/1.4NA objective lens, a colour camera and a motorized x-y sample positing stage.…”