Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Towards a surgical phase detection using Markov Logic Networks

: Philipp, P.; Bleier, J.; Fischer, Yvonne; Beyerer, Jürgen

Volltext (PDF; )

Radermacher, Klaus (Ed.) ; Society for Computer Assisted Orthopaedic Surgery -CAOS-, Bern:
CAOS 2017, 17th Annual Meeting of the International Society for Computer Assisted Orthopaedic Surgery. Papers. Online resource : June 14-17, 2017, Aachen, Germany
2017 (EPiC Series in Health Sciences 1)
International Society for Computer Assisted Orthopaedic Surgery (CAOS Annual Meeting) <17, 2017, Aachen>
Konferenzbeitrag, Elektronische Publikation
Fraunhofer IOSB ()
situation detection; surgical workflow; Markov Logic Network; machine learning; planning; interactive assistance

The use of assistance functions for diagnosis and surgical interventions has become an evolving area for mastering challenges of contemporary medicine. Inter alia, these assistance functions can help to prevent malpractices and preserve a high level of satisfaction for patients as well as employees.
To enable such functions in context of a computer-assisted orthopaedic surgery (CAOS), we elaborate the use of Markov Logic Networks (MLNs) for modelling surgical phases. In contrast to commonly researched systems for surgical process modelling, MLNs combine rule-based as well as probabilistic approaches. This allows us to integrate soft and hard constraints into our network – which greatly expands the scenery of currently researched models for phase detection in surgical interventions.
In our contribution, we present the necessary fundamentals of MLNs and show the application to a comprehensible test case. The results are promising concerning the use of MLNs for surgical phase detection. In particular, MLNs have shown two advantages: Firstly, due to their template characteristics, few logic rules allow to model numerous interdependencies between the different surgical phases. Secondly, the combination of probabilistic and logic approaches allows to handle sensor inaccuracies and misclassifications of features directly. E.g., the inaccuracy of a sensor can be expressed by reducing the weight of corresponding formulas, allowing for a softening of constraints.