Automated assessment of simulated laparoscopic surgical skill performance using deep learning

Authors: David Power, Cathy Burke, Michael G. Madden, Ihsan Ullah.

The study presents a new dataset – the Laparoscopic Surgical Performance Dataset (LSPD) – created for analyzing recordings from laparoscopic surgery simulations using artificial intelligence. Unlike previous datasets focused on tool or anatomy tracking, LSPD enables assessment of operator skills at different experience levels (novices, residents, experts) across three task types: stack, bands, and tower.

For classification, a 3D CNN neural network was applied in a weakly supervised setting, achieving high performance (F1 = 0.91, AUC = 0.92). The study demonstrates that modern computer vision methods can effectively evaluate surgical training levels without expert involvement, significantly reducing costs and increasing scalability of surgical training systems.