Using machine learning analysis to assist in differentiating between necrotizing enterocolitis and spontaneous...

Area of Science:

• Neonatal surgery
• Medical artificial intelligence
• Pediatric gastroenterology

Background:

• Necrotizing enterocolitis (NEC) and spontaneous intestinal perforation (SIP) are severe conditions in preterm infants, often necessitating surgery.
• Accurate pre-operative diagnosis is challenging without direct bowel visualization.
• Predictive analytics offer a potential solution for differentiating NEC and SIP to guide surgical management.

Purpose of the Study:

• To develop and validate machine learning models for distinguishing between NEC and SIP in neonates.
• To improve diagnostic accuracy prior to surgical intervention.
• To inform surgical decision-making in preterm infants with suspected intestinal perforation.

Main Methods:

• Analysis of patient characteristics using machine learning methodologies.
• Optimization of models based on area under the receiver operating characteristic curve (AUROC).
• Validation of the developed models using a separate patient cohort.

Main Results:

• A random forest model achieved an AUROC of 98%, outperforming a ridge logistic regression model (92% AUROC).
• The random forest model demonstrated accurate predictions when applied to the validation cohort.
• The study identified 40 patients for analysis.

Conclusions:

• Machine learning models show feasibility in differentiating NEC and SIP before surgical intervention.
• This approach can potentially enhance diagnostic accuracy and guide treatment strategies.
• Further research supports the use of AI in neonatal surgical decision-making.