Page 18 - Read Online

Page 263 Collins et al. Art Int Surg. 2025;5:254-69 https://dx.doi.org/10.20517/ais.2024.92

surgical complexity, lifestyle, genetics, and real-time recovery data to create customized recovery plans.
These plans could:
• Optimize recovery trajectories: Where AI systems propose adapting the recovery plan in real time,
[55]
ensuring pain management , physical therapy, and wound care are tailored to the patient’s progress.
• Reduce hospital stays: Where recovery metrics are monitored in real time, leading to AI systems helping
[56]
recommend when patients are ready for discharge , reducing unnecessary hospital stays while maintaining
optimal recovery outcomes.
• Offer customized exercise recommendations: Where AI systems could make personalized exercise
recommendations to strengthen abdominal muscles while ensuring the integrity of the surgical repair.

Virtual health assistants: AI is also being explored to improve patient engagement and postoperative follow-
up through automated systems and virtual health assistants. LLMs integrated into conversational AI systems
are being investigated to facilitate routine follow-ups , answer common patient concerns, and guide them
[57]
through the recovery process. AI-powered rehabilitation programs can monitor patient mobility, guide
core-strengthening exercises, and provide real-time feedback to prevent strain on the healing abdominal
wall. These interventions can be particularly beneficial for high-risk hernia repair patients who require
specialized recovery strategies.

Surgery education
The roles of AI in surgery education are likely to be wide-ranging, from realistic simulation-based
[60]
training [58,59] , creation of personalized learning pathways, and video-based competency assessment to AI-
curated case studies, technical skill assessment , and the enhancement of video-based learning platforms
[17]
such as WebSurg . Additionally, LLMs such as ChatGPT can process, generate, and synthesize vast
[61]
amounts of medical knowledge, making them increasingly valuable tools in surgery education. A very
promising future direction is the integration of LLMs with surgery video education to enable students to
learn about video content using natural conversational language. To this end, there has been some recent
research in Visual Question Answering, such as Surgical-VQA , which is an AI model trained to answer
[62]
questionnaires on surgical procedures based on the procedure’s video.

The use of LLMs regarding surgery, however, extends far beyond education, as indicated in the previous
section. The applications range from improving patient communication , helping document clinical
[63]
[64]
research , performing statistical analysis, and identifying study biases to generating code to assist the
[65]
[66]
development of medical device software .
DISCUSSION
From research to clinical deployment
Advancing from research to clinical deployment of AI in AWS
The application of AI technologies to AWS is in its infancy. As highlighted above, numerous applications
are being investigated in the research literature; however, none have yet translated into routine practice.

In the development of AI technologies for AWS and hernia repair, including our own research [16,17] , pre-
recorded surgical videos have been instrumental in model development and validation. While these
retrospective datasets provide a strong foundation, they also present key limitations.

One major challenge is that AI models are typically evaluated based on task performance metrics, such as
structure recognition rates, without direct assessment of their real-world clinical impact. Transitioning from
retrospective validation to real-world implementation in AWS is a complex, multi-step process.