Classification of Paediatric Inflammatory Bowel Disease using Machine Learning.

Area of Science:

• Computational biology
• Medical informatics
• Gastroenterology

Background:

• Paediatric inflammatory bowel disease (PIBD), encompassing Crohn's disease (CD), ulcerative colitis (UC), and IBD unclassified (IBDU), presents a growing diagnostic challenge.
• Accurate PIBD diagnosis is crucial for timely and effective treatment strategies.
• Current diagnostic methods may benefit from advanced analytical approaches to improve accuracy.

Purpose of the Study:

• To develop and validate machine learning (ML) models for classifying PIBD subtypes.
• To assess the diagnostic contribution of endoscopic and histological data, individually and combined.
• To explore ML's potential in enhancing diagnostic accuracy for PIBD.

Main Methods:

• Utilized endoscopic and histological data from 287 children diagnosed with PIBD.
• Developed and trained unsupervised and supervised ML models for disease classification.
• Validated the optimal ML model on an independent cohort of 48 PIBD patients.

Main Results:

• Unsupervised clustering revealed four novel subgroups based on colonic involvement but lacked clear CD/UC delineation.
• Supervised ML models achieved classification accuracies of 71.0% (endoscopic), 76.9% (histological), and 82.7% (combined data).
• The optimal combined model demonstrated 83.3% accuracy on an independent validation cohort.

Conclusions:

• Machine learning models, particularly when integrating both endoscopic and histological data, significantly enhance PIBD diagnostic accuracy.
• Supervised ML approaches underscore the necessity of both data types for precise PIBD subtype classification.
• This study provides a framework for applying ML to clinical data, improving diagnostic capabilities in paediatric gastroenterology.