COSHE Seminar: Jennie Karlsson

On Monday 25/11, Computational Science for Health and Environment hosts another COSHE seminar in room Gotland (F126), Geocentrum I.

The speaker is Jennie Karlsson from Mathematics. The title of her talk will be "Point-of-care ultrasound combined with deep learning for breast cancer detection in low resource settings".

Abstract:
Women in low- and middle-income countries have considerably lower chances of surviving breast cancer compared to women in high-income countries. One of the main reasons for this is the limited access to early detection. Point-of-care ultrasound (POCUS) is a cost-effective pocket-sized ultrasound device which can be used for breast examinations. POCUS combined with a deep learning (DL) based classification network has the potential to serve as a support tool for early detection of breast cancer. 

I will present three parts of this research project: managing data scarcity, out-of-distribution detection, and uncertainty quantification. To train a classification network data is needed. However, the access to point-of-care ultrasound images is limited. In the first part of the presentation, I will talk about how augmentation and cycle-consistent adversarial networks (CycleGANs) can be used to expand and diversify data. For the suggested support tool being safe in a real world setting out-of-distribution samples, such as ultrasound images of non-breast tissue, needs to be detected, this will be cover in the second part. Finally, how do we know how sure the classification network is in its prediction? This will be addressed in the last part of the presentation, where I will talk about uncertainty quantification.