Lectures and seminars Faculty event: Francisco J. Peña

Title

From Pixels to Prognosis: Scalable Self-Supervised Models for Histopathology

Abstract

In this talk, I will present self-supervised learning methods tailored for gigapixel histopathology images, with a focus on breast cancer. Traditional deep learning models rely heavily on annotated data, which is costly and time-consuming to obtain—especially in medical imaging. Our approach uses self-supervised models to learn from large-scale unlabeled data, enabling more efficient and scalable analysis. We address two major challenges: first, how to train task-agnostic models that generalize across diagnostic tasks, and second, how to process extremely large images that exceed standard hardware memory limits. Using our unique dataset of over 300,000 breast cancer images, we aim to support diagnostic, prognostic, and biomarker prediction tasks. The outcomes of this work contribute to improved clinical decision-making, more cost-effective care, and new methods in computer vision.

About the speaker

I am an assistant professor at KTH. My expertise is computer vision and unsupervised machine learning which I apply in diverse contexts ranging from digital pathology to remote sensing.

I lead the Learning Beyond Supervision Lab, where we develop AI models trained only on unlabeled, real-world data. Our work is driven by collaboration, real-world relevance, and designing methods grounded in domain knowledge. 

Faculty lunch

Join the faculty event in Wargentin! After the seminar, the faculty lunch will be served in Ljusgården. The faculty lunches are intended for employees at MEB, and a monthly net salary deduction of 260 SEK will be made for those who work here 20% or more. If you have any questions about the lunch or specific dietary needs due to allergies, etc., please contact internservice@meb.ki.se. If you won’t be participating in the lunches for all or part of the spring term, our new routine is that you need to contact hr-support@meb.ki.se at the start of a new term, to be removed from the list. Retroactive adjustments will not be possible.  

Please, check the labeling on the salads not to take special diet foods that were pre-ordered by your colleagues.