Student Seminar

Jonathan Campbell (13:00 – 13:30)
Title: Computer vision and deep learning for quantification and analysis of morphology associated with disease

Abstract: In this talk, I will present two projects. First, I will discuss how I have used computer vision to classify genetic disorders from face photographs. I will show how my model learns morphological features that are indicative of specific conditions. I will also show how I have used data augmentation to enhance the model’s generalisation, enabling it to accurately classify genetic disorders it has not been trained on.

Next, I will introduce present our hierarchical approach to tissue analysis in whole slide images of the placenta liver. Using graph neural networks, we can capture the intricate relationships between cells and their microenvironment to predict tissue types and pathologies. I will show how using data augmentation we can improve the generalisation to new institutes.

Munuse Savash (13:30 – 14:00)
Title: Precise quantification of DNA in spent culture media and its correlation to embryonic ploidy status
Abstract: Preimplantation genetic testing is a valuable tool for determining the genetic status of IVF embryos. However, the requirement for embryo biopsy necessitates specialist equipment and highly trained staff, substantially increasing costs. Additionally, there have been concerns that biopsy might risk damage to the embryo. We set out to provide the first accurate measurement of the amount of DNA in spent culture medium, and to gain an insight into why non-invasive PGT (niPGT) is less accurate than biopsy-based methods. We assessed spent media samples that underwent whole genome amplification via a novel quantitative PCR method. There was no difference in DNA quantity in media samples associated with euploid and aneuploid embryos. However, drops that held embryos until day-6 had significantly higher levels of DNA than those associated with culture to day-5 (P<0.0001). It has been hypothesised that the quantity of DNA in the medium may be predictive of the chromosomal status of an embryo. However, we found no evidence to support this notion. The amount of DNA in the medium increased with longer exposure to embryos, explaining why niPGT has higher reported accuracy when culture is extended beyond day-5.