DeepHeme: A generalizable, bone marrow classifier with hematopathologist-level performance
Morphology-based classification of cells in the bone marrow aspirate (BMA) is a key step in the diagnosis and management of hematologic malignancies. However, it is time-intensive and must be performed by expert hematopathologists and laboratory professionals. We curated a large, high-quality dataset of 41,595 hematopathologist consensus annotated single-cell images extracted from BMA whole slide images (WSIs) containing 23 morphologic classes from the clinical archives of the University of California, San Francisco. We trained a convolutional neural network, DeepHeme, to classify images in this dataset, achieving a mean area under the curve (AUC), precision and recall of 0.99, 0.89, and 0.89, respectively. DeepHeme was then externally validated on WSIs from Memorial Sloan Kettering Cancer Center, with a similar AUC, precision and recall of 0.98, 0.86, and 0.85, demonstrating robust generalization. When compared to individual hematopathologists from three different top academic medical centers, the algorithm outperformed all three. When compared with other algorithmic cell classifiers, DeepHeme can differentiate a larger number of hematopoietic cell types, while also outperforming them in F1-score, precision, and recall. Moreover, we demonstrated that DeepHeme could differentiate normal patients from those diagnosed with acute myeloid leukemia and chronic myelomonocytic leukemia, showcasing its potential for practical applications. Lastly, a web application has been developed at https://hemepath.ai/deepheme.html, allowing researchers to interact with the DeepHeme algorithm.
Objectives:
- Understand the importance of morphology-based classification of cells in the bone marrow aspirate (BMA) for the diagnosis and management of hematologic malignancies.
- Appreciate the capabilities of DeepHeme, a convolutional neural network, to classify single-cell images extracted from BMA whole slide images (WSIs) with a high degree of accuracy.
- Compare the performance of DeepHeme to individual hematopathologists and other existing methods.
Presented by:
Harry (Shenghuan) Sun
PhD Student
University of California, San Francisco
Harry Shenghuan Sun is a 4th-year Ph.D. Candidate advised by Atul J. Butte at the University of California, San Francisco, and is also a member of the Bakar Computational Health Sciences Institute. His research interests lie in artificial intelligence for computational pathology, synthetic data generation, and vision-language models.
