An Improved Endoscopic Automatic Classification Model for Gastroesophageal Reflux Disease Using Deep Learning...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Gastroenterology

Background:

• Gastroesophageal reflux disease (GERD) diagnosis relies on the Los Angeles classification for treatment.
• Artificial intelligence (AI) and deep learning show promise in assisting clinical diagnoses.

Purpose of the Study:

• To develop an AI-driven, two-stage classification model for endoscopic GERD diagnosis.
• To enhance the accuracy of GERD severity classification using deep learning and machine learning.

Main Methods:

• A two-stage approach combining transfer learning for feature extraction and machine learning for classification.
• Application of transfer learning to extract precise image features from GERD endoscopic images.
• Development of a classification model (GerdNet-RF) integrating deep learning and machine learning algorithms.

Main Results:

• The proposed GerdNet-RF model significantly outperforms previous studies in GERD endoscopic classification.
• Achieved a test accuracy of 92.5% ± 2.1%, an improvement from 78.8% ± 8.5% in prior research.
• Demonstrated enhanced automated diagnostic capabilities through AI model refinement.

Conclusions:

• The GerdNet-RF model offers a more accurate and reliable method for GERD diagnosis.
• AI-enhanced diagnostic tools can improve the assurance of patient healthcare in gastroenterology.
• This approach signifies a advancement in applying AI for precise endoscopic image analysis in GERD.