Diagnosis of Blood Cancer with Computer Vision

Next Project
A New ‚Foundation‘ for AI Models in Healthcare
Research on artificial intelligence in cancer

As of now, the evaluation of bone marrow smears by medical experts, cytomorphology, is the gold-standard in diagnosing hematologic malignancies. The way cytomorphology is performed remained unchanged and unchallenged for the last 50+ years. 

Despite its central role in the diagnostic workflow, the overall procedure is still susceptible to error. First and foremost, the evaluation of cells under the microscope is highly subjective and by no means standardized. Hence, diagnostic accuracy relies solely on the vigilant eye of the observer. Further, expert knowledge is hard to come by and it takes years to master the accurate distinction of subtle nuances present in malignant disease. What’s more, the process itself – differentiating cells, counting individual cell types, forming ratios – is entirely done by hand and thus, takes hours to days until a final report is ready. 

Diagram Diagram artificial intelligence in hematology

Therefore, we have developed a software prototype utilizing computer vision to automatically provide diagnosis from bone marrow smears. We trained our prototype on > 2000 samples of bone marrow smears from multiple centers in order to provide robust and generalizable models. So far, it accurately distinguishes between acute myeloid leukemia, acute promyelocytic leukemia, myelodysplastic syndromes and healthy controls.

Instead of tedious manual manual labour, our prototype provides a diagnosis within less than a minute with accuracies >95%. Further, our pipeline uses weakly supervised learning to identify digital biomarkers as well as molecular alterations based on visual information only. 

We are currently working on expanding our portfolio with regard to other hematological entities and biomarkers as well gathering data from large cohorts with diverse treatment regimens. 

Publications: 

Eckardt JN, Middeke JM, Riechert S, et al. Deep learning detects acute myeloid leukemia and predicts NPM1 mutation status from bone marrow smears. Leukemia. 2022;36(1):111-118. doi:10.1038/s41375-021-01408-w
https://www.nature.com/articles/s41375-021-01408-w

Eckardt JN, Schmittmann T, Riechert S, et al. Deep learning identifies Acute Promyelocytic Leukemia in bone marrow smears. BMC Cancer. 2022;22(1):201. Published 2022 Feb 22. doi:10.1186/s12885-022-09307-8
https://bmccancer.biomedcentral.com/articles/10.1186/s12885-022-09307-8

Presented at ASH2023 (Abstract Achievement Award): 
Eckardt JN, Schmittmann T, Schulze F, et al. Explainable End-to-End Supervised Learning Identifies Myelodysplastic Neoplasms in Bone Marrow Smears
https://ashpublications.org/blood/article/142/Supplement%201/2269/499767/Explainable-End-to-End-Supervised-Learning

Presented at ASH2023: 
Schneider MMK, Schulze F, Schmittmann T, et al. Automated Preselection of Regions of Interest By Deep Learning Facilitates Rapid Whole Slide Image
https://ashpublications.org/blood/article/142/Supplement%201/3662/501983/Automated-Preselection-of-Regions-of-Interest-By

 

Current Projects