Clinical Scorecard: Anatomical Site Classification in Upper Gastrointestinal Endoscopy Utilizing Deep Learning Techniques
At a Glance
Category
Detail
Condition
Upper gastrointestinal diseases including reflux oesophagitis, gastroduodenal ulcer, and early gastric cancer
Key Mechanisms
Use of oesophagogastroduodenoscopy (EGD) with AI and deep learning for anatomical site classification and quality assurance
Target Population
Patients undergoing upper GI endoscopy, especially in regions with high gastric disease incidence such as Eastern Asia
Care Setting
Endoscopy units in hospitals and clinical centers performing upper GI endoscopy
Key Highlights
Proposed a modified guideline for upper GI endoscopic photo-documentation balancing British and Japanese standards
Presented a new annotated upper GI endoscopic image dataset from routine clinical care in China to support AI development
Introduced a complete workflow including data collection, automatic ROI extraction, anatomical annotation, and deep learning classification
Guideline-Based Recommendations
Diagnosis
Perform EGD as the gold-standard procedure for diagnosing upper GI diseases and early gastric cancer
Use standardized photo-documentation of anatomical landmarks to improve lesion detection and examination completeness
Management
Follow established guidelines such as the British Society of Gastroenterology (BSG) and Japanese Systematic Screening Protocol for the Stomach (SSS) for image acquisition
Adopt the proposed modified guideline to balance detailed stomach imaging with pragmatic clinical workflow
Monitoring & Follow-up
Assess quality of photo-documentation reports based on guideline adherence to ensure comprehensive examination
Utilize AI-assisted tools to quantify and verify completeness of anatomical site coverage during EGD
Risks
Potential for missed lesions due to blind spots or incomplete anatomical site visualization during EGD
Variability in image quality and resolution across different clinical endoscopy systems may affect classification accuracy
Patient & Prescribing Data
Patients undergoing routine upper GI endoscopy in clinical settings, particularly in high gastric cancer incidence regions
Enhanced photo-documentation and AI-assisted anatomical classification may improve early detection of gastric lesions and overall diagnostic accuracy
Clinical Best Practices
Ensure comprehensive anatomical site coverage during EGD by adhering to standardized photo-documentation guidelines
Incorporate AI and deep learning tools to assist in anatomical site classification and quality assurance of endoscopic examinations
Use multi-resolution ROI extraction methods to standardize image inputs from diverse endoscopy systems
Train and validate AI models on annotated datasets reflecting routine clinical practice to improve generalizability
