Highly accurate differentiation of bone marrow cell morphologies using deep neuralnetworks on a large image dataset
Biomedical applications of deep learning algorithms rely on large, expert annotated data sets. The classification of bone marrow cell cytomorphology, an important cornerstone of hematological diagnosis, is still done manually thousands of times every day, due to a lack of datasets and trained models.
We apply convolutional neural networks (CNNs) to a large dataset of 171,374 microscopic cytological images taken from bone marrow smears of 945 patients diagnosed with a variety of hematological diseases. The dataset is the largest expert-annotated pool of bone marrow cytology images available in the literature so far. It allows us to train high-quality classifiers of leukocyte cytomorphology that identify a wide range of diagnostically relevant cell species at high precision and recall.Our CNNs outcompete previous feature-based approaches and provide a proof-of-concept to the classification problem of single bone marrow cells.
This work is a step towards automated evaluation of bone marrow cell morphology using state-of-the-art image classification algorithms. The underlying dataset represents both an educational resource as well as a reference for future AI-based approaches to bone marrow cytomorphology.