Support vector machines improve the accuracy of evaluation for the performance of laparoscopic training tasks

• DOI: 10.1007/s00464-009-0556-6
• Publication Date: 2009-06-15T20:28:22Z
• Authors: Brian Allen; Vasile Nistor; Erik Dutson; Greg Carman; Catherine Lewis; Petros Faloutsos
• Keywords: Normalization; Training set

Despite technological advances in the tracking of surgical motions, automatic evaluation laparoscopic skills remains remote. A new method is proposed that combines multiple discrete motion analysis metrics. This compared with previously metric combination methods and shown to provide greater ability for classifying novice expert surgeons.For this study, 30 participants (four experts 26 novices) performed 696 trials three training tasks: peg transfer, pass rope, cap needle. Instrument motions were recorded reduced four Three combining metrics into a prediction competency (summed-ratios, z-score normalization, support vector machine [SVM]) compared. The comparison was based on area under receiver operating characteristic curve (AUC) predictive accuracy unseen validation data set.For all tasks, SVM superior terms both AUC set. resulted AUCs 0.968, 0.952, 0.970 tasks respectively 0.958, 0.899, 0.884 next best (weighted z-normalization). correctly predicted 93.7, 91.3, 90.0% subjects' competencies, whereas weighted z-normalization 86.6, 79.3, 75.7% accurately (p < 0.002).The findings show an SVM-based provides more accurate predictions at than previous techniques. An approach should be considered computerized performance systems.