Pattern-Based Behavioural Analysis on Neurosurgical Simulation Data


Scott Buffett, Catherine Pagiatakis, Di Jiang ;
Proceedings of the 3rd Machine Learning for Healthcare Conference, PMLR 85:514-533, 2018.


This paper presents the results of an analytics-based study to determine key differences in junior resident level and expert level surgical skill when engaging with a neurosurgical simulator. Window-based time series discretization and sequential pattern analysis were used on positional data to identify frequent movement patterns and instrumentation techniques associated with each skill class. Cross-validation confirmed that a Bayesian classification model constructed using these patterns can be used to predict skill level with a high degree of confidence and accuracy on a small sample of neurosurgeons who engaged with the simulator. An analysis of movement speed also revealed that the junior residents exhibited a high degree of very slow and very fast movements, whereas the expert surgeons displayed a significantly more consistent technique of moderate-speed movements. Finally, the analysis was integrated within a cloud-based learning framework, helping to provide beneficial feedback on movement proficiency to resident surgeons in training. The presented work makes two key contributions to the field of machine learning in the medical field: the study 1) employs a low-level behaviour-based analysis of surgical technique, as opposed to high-level summary metrics such as blood loss and average force, and 2) avoids the use of expert information on neurosurgical skill within the AI engine, and thus employs an analysis that is entirely uninformed.

Related Material