Development of an artificial intelligence-based computer-aided detection system for routine gastric biopsy diagnosis

Area of Science:

• Pathology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate diagnosis of gastric biopsy specimens is crucial for patient management.
• Pathological diagnosis can be challenging due to tumor heterogeneity and subtle malignant features.

Purpose of the Study:

• To develop and validate an artificial intelligence-based computer-aided detection (CAD) system for gastric biopsy specimens.
• To assess the system's performance in detecting malignant regions and its impact on diagnostic sensitivity.

Main Methods:

• Development of an AI system trained on diverse epithelial and non-epithelial gastric tumors from a multicenter cohort.
• Application of two algorithms and three operational validity levels with optimized parameters.
• Validation using an independent dataset and a reader study to evaluate system assistance.

Main Results:

• The AI system detected malignant regions at low-magnification, integrating into pathologists' workflow.
• System assistance in a reader study was associated with improved diagnostic sensitivity.
• Small and dispersed malignant foci were identified as challenging, highlighting the system's potential benefit.

Conclusions:

• The developed AI-based pathological CAD system shows promise for aiding pathologists in routine gastric biopsy diagnosis.
• Integration of this system into clinical practice may enhance diagnostic accuracy and sensitivity.
• The system is particularly valuable for identifying subtle or scattered malignant findings.