RMDL: Recalibrated multi-instance deep learning for whole slide gastric image classification.

Area of Science:

• Digital pathology
• Computational oncology
• Machine learning in medicine

Background:

• Whole slide histopathology images (WSIs) are crucial for gastric cancer diagnosis.
• Analyzing large WSIs and identifying abnormal regions presents significant challenges in automated diagnosis.

Purpose of the Study:

• To develop an advanced multi-instance learning method for accurate gastric cancer diagnosis from WSIs.
• To address the challenges of region selection and analysis in large-scale histopathology image data.

Main Methods:

• A novel recalibrated multi-instance deep learning (RMDL) method was designed.
• The RMDL network selects discriminative instances and recalibrates their features based on learned importance coefficients.
• A large dataset of whole-slide gastric histopathology images with pixel-level annotations was created.

Main Results:

• The proposed RMDL framework achieved significant accuracy improvements compared to existing multi-instance learning methods.
• The method demonstrated effectiveness in capturing instance-wise dependencies and feature recalibration.
• Experimental validation was performed on the newly constructed gastric histopathology dataset.

Conclusions:

• The RMDL method offers a robust and accurate approach for gastric cancer diagnosis using WSIs.
• The developed framework is generalizable and can be applied to other cancer types and WSI-based diagnostic tasks.