Development of Multicenter Deep Learning Models for Predicting Surgical Complexity and Surgical Site Infection in...

Area of Science:

• Medical imaging
• Artificial intelligence in surgery
• Abdominal wall reconstruction

Background:

• Hernia recurrence and surgical site infections (SSI) are significant complications in Abdominal Wall Reconstruction (AWR).
• Accurate prediction of surgical complexity and SSI risk is crucial for patient outcomes.
• Preoperative computed tomography (CT) imaging offers potential for risk stratification.

Purpose of the Study:

• To develop and validate multicenter deep learning models (DLMs) for predicting surgical complexity and SSI risk in AWR.
• To assess the performance of DLMs using Component Separation Technique (CST) as a surrogate for surgical complexity.
• To evaluate the efficacy of DLMs in predicting SSI using preoperative CT scans.

Main Methods:

• Developed multicenter DLMs using ResNet-18 architecture with deidentified CT images from two AWR centers.
• Utilized CST as a surrogate for surgical complexity and analyzed preoperative CT for SSI risk prediction.
• Reported model performance using accuracy and Area Under the Curve (AUC).

Main Results:

• The DLM for predicting SSI demonstrated strong performance with an AUC of 0.898.
• The DLM for predicting surgical complexity using CST underperformed, achieving an AUC of 0.569.
• Patient factors significantly influenced SSI prediction, outweighing surgical practice variability.

Conclusions:

• A multicenter DLM for SSI prediction in AWR was successfully developed, showing AI's potential in risk stratification.
• The challenges in predicting surgical complexity with DLMs may stem from subjective surgical decisions and practice variations.
• AI-enhanced risk calculators hold promise for improving patient outcomes in AWR by enabling better risk stratification.