AI could help better diagnose blood diseases – archyde

How can we better diagnose blood diseases? A research group around Helmholtz Munich wants to answer this question with artificial intelligence (AI). Its aim is to facilitate the time-consuming analysis of bone marrow cells under the microscope. The researchers developed the largest open source database to date of microscopic images of bone marrow cells. You use it as the basis for an AI model with high potential for routine diagnostics.

Every day, cytologists around the world use optical microscopes to analyze and classify samples of bone marrow cells thousands of times. This method of diagnosing blood diseases was established more than 150 years ago, but it is very complex. Finding rare but diagnostically important cells is both tedious and time-consuming. Artificial intelligence has the potential to advance this method – but it requires a large amount of high-quality data to train an AI algorithm.

Largest open source database for images of bone marrow cells

The researchers at Helmholtz Munich have developed the largest open access database to date for microscopic images of bone marrow cells. The database consists of more than 170,000 individual cell images of over 900 patients with various blood diseases. It is the result of a collaboration between Helmholtz Munich, the University Hospital of the LMU Munich, the MLL Munich Leukemia Lab (one of the world’s largest diagnostic providers in this field) and the Fraunhofer Institute for Integrated Circuits.

Strengthen artificial intelligence with the help of the database

“In addition to our database, we have developed a neural network that surpasses previous machine learning algorithms for cell classification in terms of accuracy, but also in terms of generalizability,” says Christian Matek, first author of the new study. The Deep Neural Network is a machine learning concept specially developed for processing images.

The analysis of bone marrow cells has not yet been carried out with such advanced neural networks, which is also due to the fact that no high-quality, public data sets were previously available. “

Christian Matek, first author

The researchers want to expand their database on bone marrow cells in order to capture a broader range of findings and to prospectively validate their model. “The database and the model are freely available for research and training purposes – for further training of specialists or as a reference for other AI-based approaches, for example in blood cancer diagnostics,” says study director Carsten Marr.


Journal reference:

Matek, C., et al. (2021) Highly accurate differentiation of bone marrow cell morphologies using deep neural networks in a large image data set. Blood.


Leave a Reply

Your email address will not be published. Required fields are marked *