In this project, we aim to develop machine learning tools for causal discovery and causal inference, along with tools for visualizing large causal structures to a human to increase the interpretability of the structure. The developed methods will be applied for different healthcare applications, to start with, diagnostics and treatment of Achilles tendon rupture.
Besides the interaction in SeRC, the project involves tight interaction with mathematicians such as Timo Koski at KTH, Machine Learning researchers such as Cheng Zhang at Microsoft Research Cambridge, UK, and Kun Zhang at Carnegie Mellon University, USA, and medical researchers such as Paul Ackermann at Karolinska Institutet, who is a clinical expert on Achilles tendon rupture.
Causal discovery is a machine learning subject that identifies causal relationships among observed factors. Causality is to be distinguished from mere correlation, which includes causal and associative relationships between factors. Causal structures are commonly represented as graphs. These graphs are large and noisy. A navigation of these structures needs to include dynamic filtering and graph simplification. We will examine to what extent Persistent Homology can play a role in the simplification of these graphs. Furthermore, we want to examine how to incorporate user knowledge into the causal discovery by interweaving interactive graph manipulation with the automated causal discovery process.
The initial application scenario is Achilles tendon rupture treatment. Despite advancements in treatment modalities, recovery after musculoskeletal injuries, eg. tendon rupture, is still a prolonged process with very variable and often unsatisfactory outcome. Still at one year after their injury patients suffer from symptoms like pain, fatigue, and weakness. Thus, patients are often hindered to return to work. Identifying key factors causing the different outcomes after the injuries is extremely important for the treatment of such injuries. For example, identifying for which patient groups early load bearing mobilization causes better outcome is unclear but essential.