Image-based deep learning model to predict stoma-site incisional hernia in patients with temporary ileostomy: A...

Area of Science:

• Medical Imaging
• Surgical Oncology
• Predictive Analytics

Background:

• Stoma-site incisional hernia (SSIH) is a common complication after stoma creation.
• Prophylactic biological mesh implantation is used to prevent SSIH.
• Accurate prediction of SSIH risk is crucial for patient management.

Purpose of the Study:

• To develop and validate a predictive model for SSIH in patients with temporary ileostomy.
• To evaluate the performance of a novel mixed model integrating radiomics, stepwise regression, and deep learning.

Main Methods:

• Prospective establishment and validation of a mixed predictive model.
• Integration of radiomics, stepwise regression, and deep learning techniques.
• Evaluation across primary and external validation cohorts.

Main Results:

• The mixed model demonstrated superior discrimination for SSIH prediction compared to individual components.
• Area under the curve (AUC) values were 0.947 (primary), 0.876 (validation 1), and 0.776 (validation 2).
• Improved sensitivity, specificity, and precision in SSIH prediction were observed with the mixed model.

Conclusions:

• The developed mixed model offers enhanced predictive capabilities for SSIH.
• This model can aid in SSIH precaution and inform clinical decision-making.
• The integration of multiple analytical approaches provides comprehensive insights for hernia prevention.