Construction of a deep learning-based predictive model for delayed graft function in kidney transplantation.

Area of Science:

• Nephrology
• Transplant Surgery
• Artificial Intelligence in Medicine

Background:

• Delayed graft function (DGF) is a significant complication following kidney transplantation, negatively impacting long-term graft survival.
• Accurate prediction of DGF risk is crucial for optimizing patient management and improving outcomes in deceased donor kidney transplantation.

Purpose of the Study:

• To develop and validate deep learning (DL) models for predicting the risk of DGF in recipients of deceased donor kidney transplants.
• To identify the most effective DL architecture for DGF risk assessment.

Main Methods:

• Retrospective analysis of 670 deceased donor kidney transplant recipients.
• Utilized five DL algorithms: BiGRU, Convolutional bidirectional long short-term memory, convolutional gated recurrent unit, CNN-BiGRU, and CNN-bidirectional long short-term memory.
• Employed Synthetic Minority Oversampling Technique to address class imbalance and evaluated models using AUC, Matthews correlation coefficient, and F1 score.

Main Results:

• The CNN-BiGRU hybrid model achieved superior predictive performance with an AUC of 0.848.
• The model demonstrated high sensitivity (82.1%) and specificity (83.5%) in identifying DGF risk.
• The Synthetic Minority Oversampling Technique effectively managed class imbalance in the training dataset.

Conclusions:

• A CNN-BiGRU-based prediction model shows excellent performance in identifying patients at high risk for DGF.
• This AI-powered tool can aid nephrology, urology, and transplant teams in personalized risk stratification and post-transplant management.
• Implementing such models can potentially enhance patient outcomes after kidney transplantation.