Page 26 - Read Online

Page 346 Walker et al. Art Int Surg. 2025;5:345-9 https://dx.doi.org/10.20517/ais.2025.02

transforming medicine. In hernia surgery and abdominal wall reconstruction (AWR), AI has played a role
[1]
in improving surgical planning, enhancing precision during procedures, and predicting patient outcomes .
By analyzing large amounts of data and identifying patterns that might be missed or unrecognized by
humans, AI has the potential to make hernia surgery safer and more efficient. The aim of this perspective
piece is to advocate for a balanced integration of AI into hernia surgery, highlighting its benefits, challenges,
and future implications. Additionally, it emphasizes the importance of fostering collaboration with non-
medical disciplines such as computer science and data analytics to advance AI-driven surgical innovation.

The first instances of AI integration in medicine aimed to diagnose specific medical pathologies, such as
[2,3]
diabetic retinopathy and intracranial hemorrhage . Hernia surgery was the first field of medicine where AI
algorithms were used specifically to predict patient outcomes . Outcome prediction was performed with
[4]
the development of deep learning models (DLMs), a subset of AI that focuses on enabling machines to learn
from data in either a supervised or unsupervised fashion. Supervised learning utilizes known, labeled
outputs to train AI algorithms, while unsupervised learning relies on algorithms to recognize relationships
in a given dataset and train itself. Machine learning (ML) involves teaching systems to recognize patterns
and make decisions based on data, while deep learning, a more advanced subset, uses specialized neural
networks with many layers to analyze and appropriately weigh the importance of complex information .
[5]
DLMs are particularly adept at processing and analyzing visual data such as images and videos and provide
a means for objectively assessing large amounts of information. These models learn to identify patterns,
objects, or features through supervised or unsupervised learning and are trained and then tested on novel
data. Applications of these models include object detection, facial recognition, and medical imaging
analysis .
[5]

Using preoperative CT imaging, Elhage and colleagues were able to predict surgical site infections in AWR
patients with 85% accuracy and demonstrated that DLMs were more successful than expert hernia surgeons
in predicting the need for component separation techniques . Since that seminal publication, there have
[4]
been other examples whereby objective preoperative data are used to predict hernia outcomes. For example,
the models developed by Elhage were modified by Ayuso et al., who generated additional instances of “rare”
surgical outcomes to train DLMs and were able to improve the ability of models to detect instances of mesh
[1]
infection and pulmonary failure . Moreover, Lesch et al. have recently demonstrated how hernia repairs
may be improved by biomechanical analyses, such as assessments of hernia and mesh size, suture factors,
and tissue stability . AI-driven analysis of CTs that evaluate biomechanical properties of hernias may
[6]
increase the precision of hernia repair. Ultimately, the interplay between preoperative imaging and patient
data can both be used for predictive modeling, but this has not yet been done.

By analyzing large datasets of patient images and demographic information, these models can identify
patterns that indicate surgical risk and can help to effectively stratify patients. This would allow surgeons to
make tailored risk assessments and informed decisions, improving both preoperative planning and
postoperative care. Though external validation is needed before widespread use, the ultimate goal of these
AI models is to personalize the care of the hernia patient and improve the quality of life of patients, which
remains the principal goal of the hernia surgeon. The external validation of AI models, along with the
development of AI models based on multicenter data, is a requisite task before they are more broadly
generalizable . The ability to anticipate complications and surgical complexity can aid in planning the safest
[7]
surgical approach and identifying potential anatomical challenges. In addition, patients who are deemed at
particularly high risk given their comorbidities and risk stratification using AI algorithms may be triaged to
tertiary hernia centers to receive care. Patients who are at risk for a mesh infection may be followed for