OBJECTIVES:Technical perfection of performing vascular anastomosis is essential to expect excellent results after open vascular reconstructions. . The lack of both objective metrics to measure surgical skills of trainees and of the immediate feedback by qualified mentors continue to be a problem of simulation training. Machine learning methodologies offer a rapid, reproducible and scalable feedback and highly accurate clinical scoring of surgical skills, but they primary rely on video streams and kinematic data analysis lacking nuanced skills on the end-results of vascular reconstructions.The aim of this study was to propose a model that applies the method of computational fluid dynamics machine learning to give an objective and explainable assessment vascular anastomoses.
METHODS: 519 End-to-side anastomoses were executed by 119 trainees trainees during simulator-based training. We measured haemodynamic features, employing computational fluid dynamics (CFD). Multiple expert-based assessment of anastomosis structures were applied based on observation and ranking of 4-500 vascular structures. The array of algorithms encompasses linear models (Ridge, Partial Least Squares), support vector machines, and tree-based models (Random Forest, Extremely Randomized Trees, Extreme Gradient Boosting).
RESULTS: Among the suite of explored algorithms, Extreme Gradient Boosting emerged as the top performer. It attained a root mean squared error (RMSE) of 0.758 (CI95: 0.722 to 0.799) and a coefficient of determination (R2) of 0.673 (CI95: 0.617 to 0.725) with most influential parameters of most influential features are the gradient of wall shear stress (WSS), transverse WSS, and maximum helicity.
CONCLUSIONS:This investigation proposes a paradigm shift in the assessment of vascular surgical skills through the fusion of computational fluid dynamics (CFD) and machine learning (ML) approaches. Leveraging advancing computational capabilities, this methodology has the potential to supplant the labour-intensive and subjective instructor-based scoring procedures and allowing the methodology to enhance the evaluation landscape in surgical training.