Our I3lab „Model guided machine learning for simulating soft tissue materials in medicine“ was initiated as part of the TUHH I3lab program.

Physical models can predict mathematical comprehensible statements about biomedical phenomenons quickly. In view of complex tasks, they are often not as individual as the patient and are inaccurate. 

Machine learning algorithms can represent complex correlations but need large datasets for training and are hardly comprehensible.

A combination of both methods can benefit from the advantages of both approaches. The physical models can be used for an effective pre-training of the machine learning models and thereby enhance the data efficiency. The trained models can on the other hand be used to refine the physical models, which allow a formal verification of model and system characteristics.

Methods for combining models with machine learning will be investigated based on soft tissue modeling in medicine, which is an interesting materials science use case.