Mixed attention ensemble for esophageal motility disorders classification

Area of Science:

• Gastroenterology
• Medical Informatics
• Artificial Intelligence

Background:

• Esophageal motility disorders stem from lower esophageal sphincter dysfunction and peristalsis abnormalities.
• Common symptoms include dysphagia, chest pain, and heartburn.
• Current diagnosis relies on high-resolution esophageal manometry, which is complex, time-intensive, and requires specialized expertise.

Purpose of the Study:

• To develop an accurate and efficient classification model for esophageal motility disorders.
• To address the limitations of traditional diagnostic methods like manometry.

Main Methods:

• Proposed a Mixed Attention Ensemble (MAE) model integrating four base models.
• Employed a multi-dimensional attention mechanism for feature extraction.
• Utilized a mixed voting mechanism for model weighting.
• Conducted experiments with three voting strategies on a proprietary dataset.

Main Results:

• The MAE model achieved a high accuracy of 98.48%.
• Demonstrated superior performance compared to traditional voting ensembles.
• Maintained a low parameter count, indicating efficiency.

Conclusions:

• The MAE model is effective for classifying esophageal motility disorders.
• Offers a valuable tool for physician pre-diagnosis.
• Highlights the potential of AI in improving diagnostic efficiency and accuracy in gastroenterology.