Surgical Workflow Analysis: Representation and Application to Monitoring

Abstract: In modern medicine an immense amount of information has become available. Novel imaging modalities, image processing methods, electronic patient recordings and picture archiving and communication systems allow to obtain and access huge amounts of data. This development is to appreciate, but it also raises new challenges. During surgery, clinical personnel has to cope with this information overload while experiencing high stress. Often technological innovations are not well integrated into surgical workflow and they are not fully utilized. Many research focused more on technological advances than solving concrete clinical problems. For future research there is an urgent need to understand surgical workflows, integrate novel technologies better and unburden the surgeon. This work contributes to solving this problems by investigating how to represent and automatically analyze surgical workflow. While a surgery is performed, the current phase is recognized automatically. So e.g. context-sensitive user interfaces could be developed that only visualize the most important information. Other applications include automatic monitoring of the remaining information, generating human-understandable models of surgical workflow, statistical analysis for workflow optimization and automatic report generation. Hidden Markov Models (HMM) are used as a statistical model to analyze workflow based on the use of surgical instruments during a laparoscopic cholecystectomy. Different HMM-topologies and methods to build complex topologies are discussed. Their capability to represent surgical workflow and automatically recognize the current phase of a surgery is compared. Using a Markov chain the phases are recognized with an accuracy of 94.7%. Initial steps towards developing applications are taken. A graphical, human-understandable representation is generated automatically and a first impression of automatic report generation is given. It is also addressed how to predict the remaining duration of a surgery, which is an important issue for increasing the planning reliability of time schedules.