Computational Hematopathology for Improved Diagnostics

Project description

Identifying hematologic malignancies still relies on the time-consuming and subjective visual assessment of images. Every day, cytologists and pathologists are confronted with rare diagnostic cells, ever-increasing image data, and heterogeneous disease manifestations. Although we understand blood better than any other human tissue, we are unable to quantitatively predict a patient’s blood dynamics from a measurement. Diagnosis thus depends on rough staging schemes and the expertise and intuition of the clinician. In my proposal, I address these challenges by establishing computational hematopathology, a combination of artificial intelligence algorithms and mathematical models that will boost the currently prevailing manual assessment. Based on my experience in using these methods for scrutinizing stem cell differentiation I will combine the power of deep learning and mathematical modeling with digitized and expertly annotated image data. My unique approach enables me to design and parametrize a data-driven model to predict hematopoietic dynamics in health and disease. Since the interpretation of digitized slides is becoming the clinical standard, novel algorithms for standardized disease classification and improved diagnosis are critically needed now. This interdisciplinary project merges methods from digital pathology, machine learning, image processing, and mathematical modeling. ComHematoPathology will provide novel approaches and software tools for automated classification of hematopathology image data, allowing for reproducible and precise diagnosis at an unprecedented level. This will increase throughput and standardize the diagnosis of blood diseases and will thus improve the treatment of patients suffering from hematologic malignancies.