Bayesian Deep Learning for Opportunistic Disease Screening
Many medical images are acquired for some primary purpose, such as identifying the sources of abdominal or chest pain, as well screening for diabetic retinopathy. For the first purpose, it is common to use abdominal Computer Tomography (CTA), for the second – chest X-ray or CT, and for the third – retinal imaging modalities. What unites all of them, is that the associated images are high in volume (collected routinely), and thanks to AI methods, they can reveal a lot of additional information about the patient, and especially help to opportunistically identify subjects at risk of future, potentially deadly diseases.
Predicting the incidence of comorbidities, such as stroke, is of high value, and it could shift the healthcare system from reactive to proactive. While there are these attractive opportunities, even the state-of-the-art AI methods lack capabilities to estimate uncertainty, which can mislead clinicians who use these models. In our project, we aim to tackle this challenge and bring computationally efficient Bayesian Deep Learning methods to the opportunistic disease screening domain.
Researchers working in the project
Doctoral researcher Helinä Heino, MSc
Post-doctoral researcher Egor Panfilov, MSc (PhD 2023)
Post-doctoral researcher Huy Hoang Nguyen, MSc (PhD 2023)