Automated Machine Learning Pipeline Framework for Classification of Pediatric Functional Nausea Using High-Resolution...

Area of Science:

• Gastroenterology
• Biomedical Engineering
• Computational Biology

Background:

• Pediatric functional nausea lacks objective diagnostic tools, complicating patient management and clinical treatment.
• Developing a non-invasive screening tool is crucial for accurate diagnosis and effective treatment strategies.

Purpose of the Study:

• To present an innovative method for objectively classifying pediatric functional nausea.
• To utilize cutaneous high-resolution electrogastrogram (hr-EGG) data for this classification.

Main Methods:

• An Automated Electrogastrogram Data Analytics Pipeline was developed.
• A 3x8 factorial design was employed to identify an optimal classification model.
• Synthetic hr-EGG data were used to validate the pipeline and assess noise reduction techniques.

Main Results:

• A 10-parameter support vector machine (SVM) binary classifier with a radial basis function kernel was identified as the top-performing model.
• The optimal model achieved a 91.6% test Receiver Operating Characteristic Area Under the Curve (ROC AUC) score.
• The pipeline demonstrated clinically significant objective classification performance.

Conclusions:

• Automated machine learning analysis of hr-EGG data enables objective classification of pediatric functional nausea.
• This approach shows potential as a data-driven screening tool for objective diagnosis.
• This is the first study to report clinically significant objective classification of pediatric nausea using hr-EGG data.