An interpretable deep learning framework for intestinal metaplasia detection in gastric histopathology images

Category	Detail
Condition	Intestinal Metaplasia
Key Mechanisms	Deep learning-based automated detection using ConvNeXt-Tiny architecture with Generalized Mean pooling and Efficient Channel Attention.
Target Population	Patients with gastric biopsy samples indicating intestinal metaplasia.
Care Setting	Histopathological assessment in clinical pathology laboratories.

CNXTGeM achieved 99.04% accuracy and 100% sensitivity in detecting intestinal metaplasia.
The model outperformed baseline deep learning models including VGG16 and DenseNet121.
External validation on the GasHisSDB dataset showed robust performance with 99.34% accuracy.
Gradient-based visualization techniques confirmed model focus on relevant histopathological features.
The framework may reduce inter-observer variability in histopathological assessments.

Utilize hematoxylin and eosin staining for initial assessment of intestinal metaplasia.
Consider immunohistochemical markers such as MUC2, CDX2, and GATA4 for diagnostic support.

Implement computer-assisted detection tools to enhance diagnostic accuracy and workflow efficiency.

Regular surveillance for patients with diagnosed intestinal metaplasia due to its precancerous nature.

Delayed identification of intestinal metaplasia may increase the risk of progression to gastric cancer.

Individuals with chronic gastric inflammation and suspected intestinal metaplasia.

Early identification and monitoring of intestinal metaplasia are crucial to prevent progression to gastric cancer.

Incorporate AI-based tools in histopathological workflows to improve diagnostic accuracy.
Ensure thorough training and validation of AI models with diverse datasets to enhance generalizability.

Clinical Scorecard: A Deep Learning Approach for the Automated Identification of Intestinal Metaplasia in Gastric Histopathology Images