Reassessment of deep learning-based surgical phase recognition in laparoscopic cholecystectomy

Dear Editor,

We have read with great interest the research article “Deep learning-based surgical phase recognition in laparoscopic cholecystectomy published in volume 28, Ann Hepatobiliary Pancreat Surg [1]. This utilitarian research demands investigating and describing a deep learning model created to spontaneously predefine surgical applications in laparoscopic cholecystectomy via a publicly available dataset consisting of practical videos from the Cholec80 dataset and 40 institutional videos about artificial intelligence (AI). For this purpose, Yang et al. [1] divided the data into two: i) training sets and ii) testing sets at a ratio of 2:1. Herewith, more extensive sampling instead of a limited dataset, which might not sufficiently obtain the diversity of performed surgical techniques and performing surgeons, might lead to poor model performance in original situations, distort, or permute to the aftermath(s) of this AI work. In addition, a feasible execution of real-time system(s), per se, could alter the premise of this study. As such, future multicenter AI research from various institutions focusing on broader sampling in order to contribute to resolving this pathology might augment the resilience of the model and its serviceability in multiple surgical procedures. Last but not least, discovering unsupervised or semi-supervised deep learning-based methods could induce a tremendous amount of unannotated surgical data to advance model training without comprehensive manual annotation. Finally, real-time assessment tools during the surgical procedure are essential to ensure rapid feedback, enabling better training, boosting predictive capabilities, and, all in all, ameliorating surgical outcomes. This issue merits further investigation. We thank Yang et al. [1] for their valuable study on AI in Ann Hepatobiliary Pancreat Surg—post tenebras lux.

Funding Statement

FUNDING None.