Artificial intelligence and robot-assisted surgery: The use of deep learning in surgical data science

Abstract

Robotic-assisted surgery (RAS) has transformed minimally invasive surgery by offering improved precision, dexterity, and ergonomics. However, even with these technological advances, surgical outcomes still depend on surgeon skills, making reliable and objective performance assessment during training important.<br/>Traditional evaluation methods rely on expert reviewers or structured scoring systems, which are often time-consuming, inconsistent, and difficult to scale. Artificial intelligence (AI), particularly deep learning (DL), offers a promising alternative by enabling automated analysis of data produced during RAS procedures. However, progress has been limited by data scarcity, inconsistent annotation practices, and the complexity of developing models that generalize across different surgical settings.<br/>This thesis bridges the gap between AI and RAS by developing and evaluating DL approaches capable of automated video-based surgical assessment. It introduces open, standardized methods for collecting and preparing surgical video data, proposes model architectures for action recognition and skill assessment, and demonstrates how transfer learning can overcome the challenges of small clinical datasets. <br/>The work shows that DL models trained on video-based diverse datasets can assess surgical performance and adapt to new procedures and clinical environments.<br/>

Ähnliche Arbeiten