Agenda

"We hear form the ECLT network" series:

"Probabilistic Graphical Models Turn Clinical Decision Support Explainable"

Sven Hirsch
Head the Center of Computational Health at the school of Life Sciences and Facility Management of ZHAW

17th February 2023, 3.00 PM CET

>> To participate via ZOOM, please use this link.

Abstract:
Probabilistic Graphical Models Turn Clinical Decision Support Explainable
The central pillar of evidence-based medicine is to create validated models from unbiased observation. As in most cases it isn’t possible to conduct randomized control studies, we need to find ways to build data driven models from heterogeneous, possibly biased data. We exemplify this challenge on intracranial aneurysms where clinicians struggle to formalize the risk of intracranial aneurysm rupture. We show that Bayesian networks (BNs) provide insights into the mechanistic rupture risk factor relationships, while accounting for interactions between all factors. A better understanding of the complex interplay of factors can guide treatment decisions. Modelled by de-confounding the variables, BNs reconstruct the graphical representation of the joint distribution of the data-generating structural model and include expert knowledge. The models perform superior to factor models and are on par with machine learning approaches. Interpretability is a clear advantage of BNs, which is key for clinical adoption of decision-making tools. In our example BN models confirmed the findings from standard analysis methods and uncovered additional associations in the data.

Short bio
Prof. Sven Hirsch, PhD
Sven head the Center of Computational Health at the school of Life Sciences and Facility Management of ZHAW.  He is a physicist, active in the field of computational health, and holds a post-gradual degree in media arts. In his work he merges statistical approaches like machine learning with mechanistic modelling to reproduce disease mechanisms and clinical pathways. He searches for new disease biomarkers from clinical images, time series signals, and patient data to improve diagnosis and offer decision support. The portfolio of the research center encompasses biosensor analysis and digital health, medical data modelling, medical image analysis and biomedical simulation. The center embraces applied research questions and collaborates with industry, hospitals to deliver innovative digital health solutions and to translate computational health research into clinical practice. Sven directs the ZHAW digital health lab, an interdisciplinary platform that innovates patient-centered data-driven health care.