Artificial intelligence aids doctors in diagnosing necrotizing enterocolitis and predicting surgery using abdominal...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neonatal Surgery

Background:

• Neonatal necrotizing enterocolitis (NEC) diagnosis is challenging due to subtle radiographic signs and interpreter variability.
• Delayed diagnosis of NEC can lead to adverse outcomes in preterm infants.
• Accurate interpretation of abdominal radiographs (ARs) is crucial for timely NEC management.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) based artificial intelligence (AI) model for predicting NEC from ARs.
• To assess the AI model's ability to differentiate between medical and surgical NEC cases.
• To determine if AI assistance improves clinician accuracy in interpreting ARs for NEC diagnosis.

Main Methods:

• Retrospective collection of 738 ARs from 576 preterm infants across three centers.
• Training and internal testing of six deep learning models (Efficientnet, Inception_v3, VGG, Resnet, Squeezenet, Densenet) on data from two centers.
• External validation on a third center's data, comparing clinician accuracy with and without AI-assisted visualizations.

Main Results:

• The Efficientnet-b0 model achieved high AUC values on external test data (0.883 for non-NEC, 0.640 for medical-NEC, 0.837 for surgical-NEC).
• The AI model demonstrated superior performance compared to single-center models.
• Clinician diagnostic accuracy improved by 2.0% to 26.2% with AI assistance.

Conclusions:

• A CNN-based AI model can effectively predict NEC and identify surgical intervention needs using ARs.
• The developed AI model shows promise in supporting clinical decision-making for NEC.
• AI assistance enhances the accuracy of interpreting abdominal radiographs for neonatal necrotizing enterocolitis.