Abstract
INTRODUCTION: The clinical diagnosis of hematological diseases depends on the differential count of nucleated cells on the bone marrow (BM) smears, and an artificial intelligence (AI)-based system was applied to automatically classify BM nucleated cells in pediatric hematological disease samples in this study. METHODS: The BM aspirate smears were collected from 213 pediatric patients (under 18 years old) at Harbin Medical University Affiliated Sixth Hospital from October 2023 to June 2024. The entire smear of BM was scanned by a ×40 objective lens to obtain complete digital images using an automated analysis method named Morphogo. Next, Morphogo was used to capture nucleated cells in an area of BM smears that was selected by hematopathologists, with a magnification of ×100 objective lens. RESULTS: Morphogo demonstrated a high overall accuracy (> 87.8%) in pre-classifying nucleated cells in BM aspirate smears obtained from masked pediatric patients. In addition, the average values of sensitivity and accuracy in Morphogo cell classification were remarkably high. Moreover, Morphogo could reduce the time costs on classifying BM nucleated cells. Besides, there were positive correlations between Morphogo and manual categorization for immunologic thrombocytopenic purpura, BM failure, hyperplastic anemia, acute leukemia, chronic myeloid leukemia, and other hematological diseases. CONCLUSION: This research demonstrated the clinical potential of the Morphogo in early screening of pediatric hematological diseases and its reliability as an automated tool for differential counting and analysis of BM nucleated cells.