Table 2 Area under the curve values of prediction models for postoperative pancreatic fistula.

From: Deep learning-based prediction of post-pancreaticoduodenectomy pancreatic fistula

 

Prediction models

Training set

Validation set

Test set

For postoperative pancreatic fistula

Roberts model

0.662

0.731

0.637

ML model

0.744

0.769

0.730

DL model

0.859

0.745

0.714

Ensemble model

0.969

0.779

0.750

For clinically relevant postoperative pancreatic fistula

Roberts model

0.647

0.623

0.635

ML model

0.710

0.785

0.623

DL model

0.978

0.717

0.622

Ensemble model

0.936

0.915

0.682

  1. For the ML and DL models, values of the single model which showed the best predictive performance are shown.
  2. ML clinical and body composition data-based machine learning model, DL computed tomography-based deep learning model.
Back to article page