Machine Learning Classification of Inflammatory Bowel Disease in Children Based on a Large Real-World Pediatric...

Area of Science:

• Computational medicine
• Pediatric gastroenterology
• Machine learning applications

Background:

• Rising incidence of pediatric inflammatory bowel diseases (PIBD) necessitates improved diagnostic methods.
• Existing machine learning models for PIBD classification often rely on complex data like histology or extensive serology.
• There is a need for diagnostic tools utilizing promptly available clinical data for enhanced patient care.

Purpose of the Study:

• To evaluate the performance of various machine learning algorithms for classifying PIBD using readily accessible data.
• To assess the diagnostic accuracy of models trained on initial and follow-up patient visit data.
• To optimize the best-performing algorithm for clinical application in PIBD diagnostics.

Main Methods:

• Data from the CEDATA-GPGE registry, including inflammatory bowel conditions from initial and follow-up visits, were utilized.
• Patients in remission were excluded, and categorical data were numerically encoded.
• Support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), dense neural network (DNN), and convolutional neural network (CNN) models were employed and compared.

Main Results:

• An optimized convolutional neural network (CNN) achieved up to 90.57% accuracy on data from 2018.
• Other models showed varying performance, with DNN reaching 88.78% and XGBoost at 83.94%.
• Prediction accuracy was higher for more recent datasets (2018) compared to older ones.

Conclusions:

• Machine learning classification, especially using CNNs, is a viable approach for PIBD diagnostics with timely available data.
• The study demonstrates the potential of AI in improving diagnostic latency and quality of care for PIBD.
• Further optimization of CNNs can enhance the accuracy and clinical utility of PIBD diagnostic tools.