Automated Chicago Classification for Esophageal Motility Disorder Diagnosis Using Machine Learning.

Area of Science:

• Medical imaging analysis
• Machine learning in gastroenterology
• Computational diagnostics

Background:

• Esophageal motility disorders require accurate diagnosis using the Chicago Classification algorithm.
• Current diagnostic processes can be labor-intensive and require expert interpretation.
• Automating this classification could improve efficiency and consistency.

Purpose of the Study:

• To develop and validate a machine learning-based solution for automating the Chicago Classification algorithm.
• To integrate deep learning models for precise identification and classification of esophageal motility diseases.
• To achieve automated diagnosis from image preprocessing to final classification.

Main Methods:

• Image preprocessing involved locating the swallowing instant, resizing, and rescaling.
• The InceptionV3 deep learning model was used for precise Integrated Relaxation Pressure (IRP) class identification.
• The DenseNet201 CNN architecture classified images into 5 categories of swallowing disorders.
• A combined approach of two machine learning models automated the Chicago Classification.

Main Results:

• The automated system achieved a top-1 accuracy of 86%.
• The system also obtained an F1-score of 86%.
• The entire workflow, from preprocessing to diagnosis, was automated with no human intervention.

Conclusions:

• Machine learning, specifically deep learning models like InceptionV3 and DenseNet201, can effectively automate the Chicago Classification algorithm.
• This automated approach offers a high degree of accuracy and efficiency for diagnosing esophageal motility diseases.
• The developed solution has the potential to streamline clinical workflows and improve diagnostic consistency.