Synthetic video creation of surgical tasks using generative AI for surgical training

Abstract

Abstract Introduction The aim of this study is to develop synthetic surgical training videos using generative artificial intelligence (AI) and to evaluate their realism and formative application. Material and Methods The Cogvideo X tool was used to develop the videos using generative AI. To evaluate their realism and formative use, 6 videos of different surgical tasks (knot tying, suturing and needle passing) were obtained: 3 real and 3 synthetic videos. The real videos were obtained from the JIGSAWS public dataset for surgical skills assessment. The videos were randomly evaluated by 11 experts in surgical training. Sensitivity (SENS), specifivity (SPEC), false positive rate (FOUTR), false negative rate (ER), critical success index (CSI), accuracy (ACCU) and the F1 Score (F1Score) were evaluated to measure the quality of the results. Results The results of this study were: SENS: 0.8485, SPEC: 0.8182, FOUTR: 0.1818, ER: 0.1515, CSI: 0.7179, ACCU: 0.8333 and F1Score: 0.8358. These results evidence the experts’ ability to distinguish between real and synthetic videos. The usefulness of AI-generated videos was rated by the experts with a score of 5 out of 5, which positively outstanding their potential and usefulness for surgical training and education. Conclusions Generative AI using the Cogvideo X tool enables the design of specific surgical training activities with several medical conditions and clinical scenarios. Its application was positively validated by multiple experts, who indicate its potential in surgical training due to its realism.

Ähnliche Arbeiten