Development of Machine-Assisted, Human-Centred Bone Marrow Cell Classification: Feasibility Analysis in Patients With Myelodysplastic Syndromes.

Background: Cytomorphological examination and classification of bone marrow (BM) cells using a microscope is essential for diagnosing various diseases affecting the haematopoietic system. However, this requires expertise, is effort-intensive and is inherently subjective. Methods: We developed a method for automatically capturing BM cell images at the single-cell level using an existing cell image analyser designed for peripheral blood. The captured BM cell images were pre-classified by the same analyser and subsequently viewed on a large screen by cytomorphologists for BM cell classification. The data consistency in BM blast percentages between the digital and conventional optical microscopy (CM) methods was examined. Results: During the 3.5-year study period, 657 (96.2%) of the 683 aspirated samples underwent BM cell classification using the digital method. Digital cell images were captured at 2000× magnification and stored permanently, which allowed high-quality cell image analysis at any time by anyone. Cell images could be compared side-by-side across different cell classes, which improves the cell-classification accuracy. The detailed cell morphology in some cases differed slightly between the digital images and CM. BM blast percentage was comparable between the developed method and CM when examining patients with myelodysplastic syndromes (MDS) and other conditions that need to be differentiated from MDS. In a multicentre study, cytomorphologists who used this method for the first time could perform cell classification. Conclusion: This machine-assisted and human-centred approach enhances objectivity in BM cell classification, aligns with the needs of the digital era and facilitates data sharing.