The research project involves the company Vitrolife A/S, specialising in assisted reproductive technology, and the industrial postdoc at the Department of Engineering, Aarhus University, Mikkel Fly Kragh, working on modern machine learning technologies applied on time-lapse microscopy imaging.
In vitro fertilisation (IVF) treatment is a billion dollar industry. The treatment is performed by fertilising a number of eggs by sperm (producing embryos) outside the body where they are cultured for 2-6 days. Finally, one or more embryos are transferred to the mother’s uterus with the aim of establishing a successful pregnancy. The main challenge is to maximise the probability of pregnancy by choosing the most viable embryo(s) for transfer and for potential cryopreservation (freezing). Today, this is done manually, resulting in tedious work and subjective assessments.
The current project investigates modern computer vision and deep learning technology on time-lapse Hoffman modulation contrast (HMC) microscopy imaging. The objectives are to improve clinical workflow and provide objective and possibly new and undiscovered measures of embryo viability directly related to the probability of pregnancy. Although HMC microscopy remains the preferred imaging modality among embryologists, there only exist a few prototype software systems for automated analysis of this type of image data. The project seeks to transfer recent breakthroughs within computer vision and deep learning to a promising but relatively unexplored field.