Background

There lies a tremendous but unmet potential in digitalisation of operating rooms or interventional suites in terms of workflow analysis. A detailed analysis of the complex processes can provide valuable insights about typical sub-procedure lengths, standard process patterns or individual deviations from it. As an example, we propose a novel process model for mitral valve repair to enable visualization and quantification of differences in workflow over a cohort of patients.

Methods

From 01/2020 to 01/2022, 30 patients were retrospectively included in the study (23 male, 7 female; average age 59,8±14 years; mean BMI of 26,10±4,54). All of them suffered from high-grade mitral valve insufficiency and received an endoscopic mitral valve repair. 5 different surgeons performed the procedures. The endoscopic data stream was recorded and annotated with the open-source tool ANVIL according to the pre-defined taxonomy, i.e., each frame was assigned to a temporal segment. Our hierarchical taxonomy divides the procedure into broader steps such as “chordae implantation “, “annuloplasty ring implantation”, “saline test”, and finer tasks. During the annotation process, each frame is assigned to a specific (broader) step and (more detailed) task. The data is then loaded into a novel web-based tool (Fig.1) that can visualise and quantify certain interesting metrics, such as the length of each step/task, the order in which the steps/tasks were performed and the number of changes between them. Data were analysed by splitting the cohort into subcollections according to specific criteria. We aim to present one example of a single surgery compared to the aggregation of all others in the following.

Results

Our taxonomy contains seven steps and 34 tasks, from the beginning of the incision until closing. In this work, we focussed on annotation of four “core” steps (Fig.2). In total, 1006 different segments of different tasks and steps were annotated over all 30 procedures. Interestingly, a higher number of task changes was associated with a longer procedure length (r=0,79). Moreover, the number of saline tests performed also correlated with procedure length (r=0,69). Procedures with four or less saline test performed were significantly shorter than procedures with 5 or more (p<0,05).

To provide a more detailed example, we identified the surgery where most changes in steps/tasks occurred (in total 28) and compared it against the rest of the group (13.4±5.7 changes on average) (Fig.3).The comparison showed that the surgeon performed saline tests on 13 different instances while the average amount in the rest of the cohort was 5,0±3,0. This is the highest value for this step in the collective. Furthermore, the chordae implantation lasted 55 min over a 32±16.4 min average. In addition, the alternation between implantation of the chordae and saline test shows that the surgeon had to interrupt the implantation 12 times to check the coaptation of the valve. Therefore, it can be assumed that chordae implantation represented a particular challenge in that procedure.

Conclusion

Physicians are offered a novel opportunity to review their own workflow patterns and compare them to those of others, which will become crucial for expert-driven procedures like mitral valve repair. In general, the tool provides means for future standardisation of procedures and will certainly become relevant for the entire staff to support surgical and interventional management tasks.