Artificial intelligence-powered early identification of refractory constipation in children

Area of Science:

• Pediatric Gastroenterology
• Medical Imaging Analysis
• Machine Learning in Healthcare

Background:

• Refractory constipation in children leads to severe symptoms and reduced quality of life, persisting into adulthood.
• Early identification of refractory constipation is crucial for effective management.
• Understanding colonic anatomical differences in children with constipation types is essential.

Purpose of the Study:

• To characterize colonic anatomy in children with functional constipation, refractory constipation, and no constipation.
• To develop a supervised machine learning model for early identification of refractory constipation in children.

Main Methods:

• Retrospective study analyzing patient characteristics and standardized colon size (SCS) ratios from barium enema (BE) in three groups: functional constipation (n=77), refractory constipation (n=63), and non-constipation (n=65).
• Statistical analyses to identify significant differences in colonic dimensions.
• Development of a supervised machine learning model using SCS ratios for classification.

Main Results:

• Significant differences were observed in rectum diameter, sigmoid diameter, descending colon diameter, transverse colon diameter, and rectosigmoid length across the three groups.
• A linear support vector machine model, using five SCS ratios (sigmoid colon, descending colon, transverse colon, rectum, and rectosigmoid), achieved 81% accuracy in classification.
• The model's 95% confidence interval for accuracy was 79.17% to 83.19%.

Conclusions:

• A supervised machine learning strategy demonstrated 81% accuracy in distinguishing children with refractory constipation.
• Combining barium enema imaging with a machine learning model offers practical utility for guiding the management of pediatric refractory constipation.
• This approach holds significant potential for improving clinical decision-making in pediatric constipation cases.