Clinically applicable histopathological diagnosis system for gastric cancer detection using deep learning

Area of Science:

• Digital pathology
• Computational pathology
• Artificial intelligence in oncology

Background:

• Early detection and accurate histopathological diagnosis are crucial for successful gastric cancer treatment.
• A global shortage of pathologists necessitates innovative solutions to manage workload and enhance diagnostic precision.
• Artificial intelligence (AI) assistance systems present a promising approach to support pathologists.

Purpose of the Study:

• To develop and evaluate a clinically applicable AI system for assisting in the histopathological diagnosis of gastric cancer.
• To assess the diagnostic performance of the AI system on a large, real-world dataset and external validation sets.

Main Methods:

• Development of a deep convolutional neural network (CNN) model.
• Training the CNN using 2,123 pixel-level annotated H&E-stained whole slide images (WSIs).
• Validation of the model on a test dataset of 3,212 WSIs from three scanners and an additional 1,582 WSIs from two other medical centers.

Main Results:

• The AI system achieved near 100% sensitivity and an average specificity of 80.6% on the real-world test dataset.
• The system demonstrated robust performance across WSIs from different scanners and medical centers.
• The AI system has the potential to improve diagnostic accuracy and reduce misdiagnoses.

Conclusions:

• The developed AI assistance system is feasible for routine clinical practice in histopathological diagnosis of gastric cancer.
• AI systems can effectively alleviate pathologist workload and enhance diagnostic accuracy.
• This technology offers significant benefits for improving gastric cancer detection and patient outcomes.