Leveraging machine learning to identify blood cancer types

Diagnosing the various types of blood cancer at an early stage correctly proves to be rather difficult even for trained experts. However, the disease subtypes may lead to very different clinical outcomes, some of which may be mild whereas others can lead to fatal conditions such as acute leukaemia. Thus, a timely and correct diagnosis is crucial for treating the patient appropriately. The histopathological assessment of bone marrow biopsies is a central part of the diagnostic process, but remains heavily constrained by the reliance on subjective, qualitative and poorly reproducible criteria. Computational methods leveraging recent advances in computer vision and deep learning have the potential to transform the current clinical gold standard into a reproducible and quantitative method. First, I will introduce the disease and its challenges with respect to image analysis. Then, I will describe our framework for characterising the cell population which drives the disease and its cellular microenvironment.