Schedule for the session
1st presenter - Andreas Kleppe, OUS
Time
From 10:00 to 11:00
Title
“Training better computational pathology models to improve cancer care”
Abstract
Histopathology is the gold standard of diagnosis of cancer and many other diseases. The pathology departments at hospitals in Norway and many other countries has just become digital in order to facilitate usage of AI in routine clinical practice. Computational pathology has attracted significant academic and commercial interest the last decade but entails challenges related to large image size and generalisation issues. In this talk, I will present approaches to mitigate these challenges and exemplify how computational pathology may improve cancer care.
2nd presenter - Matt Salganik, Princeton University
Time
From 11:00 to 12:00
Title
“The life course perspective for precision health: New opportunities from population registries, biobanks, and AI”
Abstract
For over a century, life course researchers in the social sciences have sought to understand how individual life trajectories evolve over time, across domains, and in context — but data and methodological limitations have constrained progress. Today, population-scale registries and new machine learning methods for sequences offer the potential to overcome these constraints and create opportunities for collaboration between social scientists and biomedical researchers. I'll describe two works in progress using rich, population-scale registry data from the Netherlands and inspired by new AI methods. First, to study how life trajectories evolve, we develop the "book of life" approach, which represents life course data as a textual narrative - rather than as a vector of numbers. Then, we fine-tune an open-weight large-language model (LLM) on millions of these narratives in order to predict life outcomes, such as having a child. This approach opens new possibilities in life course research, including the integration of environmental, social, medical, and biological data. Second, we ask: how much do birth circumstances constrain life outcomes? Using 200,000 twin pairs who share birth circumstances — everything measured and unmeasured about family, social position, and historical timing — we estimate how much of life is determined at birth versus shaped by what comes after. This approach also reveals fundamental limits to prediction, no matter how sophisticated the model or complete the data. I'll conclude by discussing opportunities for collaboration in settings where registries and biobanks enable integration of genomic, social, and medical data across the life course.
BiLS chair
Jonas Paulsen