A deep learning model for gastric diffuse-type adenocarcinoma classification in whole slide images

Area of Science:

• Gastroenterology
• Computational Pathology
• Oncology

Background:

• Gastric diffuse-type adenocarcinoma is increasingly diagnosed in younger individuals, presenting with poorer prognosis than intestinal type.
• Distinguishing diffuse-type adenocarcinoma from non-neoplastic lesions like gastritis is diagnostically challenging due to subtle cellular morphology.
• Accurate diagnosis is crucial for appropriate patient management and treatment strategies.

Purpose of the Study:

• To develop and evaluate deep learning models for classifying gastric diffuse-type adenocarcinoma from whole slide images (WSIs).
• To assess the diagnostic performance of AI models across diverse datasets.

Main Methods:

• Training of deep learning algorithms on WSIs of gastric tissue.
• Validation of models using five independent test sets from various sources.
• Performance evaluation using receiver operator curve (ROC) analysis and area under the curves (AUCs).

Main Results:

• Deep learning models achieved high diagnostic accuracy, with ROC AUCs ranging from 0.95 to 0.99.
• Consistent performance was observed across multiple, distinct test datasets.
• The models demonstrated robust capability in identifying gastric diffuse-type adenocarcinoma.

Conclusions:

• AI-based computational pathology holds significant potential for assisting pathologists in diagnosing gastric diffuse-type adenocarcinoma.
• The developed deep learning models can serve as a valuable tool in the diagnostic workflow, potentially improving efficiency and accuracy.
• Further integration of AI in pathology could enhance the early detection and management of gastric cancer.