Skip to main content
. 2023 Mar 30;7(6):780–796. doi: 10.1038/s41551-023-01010-8

Fig. 4. Decoding surgical skills from videos and simultaneous provision of reasoning.

Fig. 4

a,b, We train SAIS on video samples exclusively from USC to decode the skill-level of needle handling (a) and needle driving (b), and deploy it on video samples from USC, SAH and HMH. Results are an average (±1 standard deviation) of ten Monte Carlo cross-validation steps. c,d, We also present the attention placed on frames by SAIS for a video sample of low-skill needle handling (c) and needle driving (d). Images with an orange bounding box indicate that SAIS places the highest attention on frames depicting visual states consistent with the respective skill assessment criteria. These criteria correspond to needle repositions and needle adjustments, respectively. e, Surgical skills profile depicting the skill assessment of needle handling and needle driving from a single surgical case at SAH. f,g, Ratio of low-skill needle handling (f) and needle driving (g) in each of the 30 surgical cases at SAH. The horizontal dashed lines represent the average ratio of low-skill activity at USC.