Identification of Mutations in Giemsa-Stained Bone Marrow Images of Acute Myeloid Leukemia

In this study, we propose a robust methodology for identifica- tion of myeloid blasts followed by prediction of genetic muta- tion in single-cell images of blasts, tackling challenges associ- ated with label accuracy and data noise. We trained an initial binary classifier to distinguish between leukemic (blasts) and non-leukemic cells images, achieving 90% accuracy. To test the model’s generalization, we applied this model to a sepa- rate large unlabeled dataset and validated the predictions with two hemato-pathologists, finding an approximate error rate of 20% in the leukemic and non-leukemic labels. Assuming this level of label noise, we further trained a four-class model on images predicted as blasts to classify specific mutations. The mutation labels were known for only a bag of cell im- ages extracted from a single slide. Despite the tumor label noise, our mutation classification model achieved 85% accu- racy across four mutation classes, demonstrating resilience to label inconsistencies. This study highlights the capability of machine learning models to work with noisy labels effectively while providing accurate, clinically relevant mutation predic- tions, which is promising for diagnostic applications in areas such as hemato-pathology.