Diagnosing Necrotizing Enterocolitis via Fine-Grained Visual Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neonatal Medicine

Background:

• Necrotizing Enterocolitis (NEC) is a severe intestinal disease in premature infants.
• Abdominal X-rays (AXRs) are crucial for NEC diagnosis but present interpretation challenges.
• Accurate and timely diagnosis is vital for effective NEC management and treatment decisions.

Purpose of the Study:

• To develop and evaluate AIDNEC, a deep learning model for automated NEC detection and severity stratification from AXRs.
• To improve the accuracy and efficiency of NEC diagnosis in neonates.
• To provide a tool that assists clinicians in differentiating NEC from other conditions and determining appropriate treatment.

Main Methods:

• Developed AIDNEC, a deep learning model integrating Detection Transformer and Graph Convolution modules.
• Employed Fine-Grained Visual Classification (FGVC) by combining local and global image features.
• Trained and evaluated the model on a dataset of 1153 AXRs from 334 patients with confirmed NEC or no pathology.

Main Results:

• AIDNEC achieved 79.7% accuracy in classifying NEC versus no pathology.
• The model demonstrated statistically significant improvements over its backbone, FGVC models, and CheXNet.
• AIDNEC successfully identified discriminative regions in AXRs, supporting its classification decisions.

Conclusions:

• AIDNEC offers a robust and accurate AI-driven solution for NEC detection and severity assessment in neonatal AXRs.
• The model's performance suggests potential for widespread clinical adoption to enhance diagnostic capabilities.
• Further validation on diverse datasets confirmed AIDNEC's generalizability and robustness in medical image analysis.