Toward Robust Histology-Prior Embedding for Endomicroscopy Image Classification.

Area of Science:

• Medical image analysis
• Computer-aided diagnosis
• Representation learning

Background:

• Learning discriminative features is crucial for medical image analysis but is hindered by small datasets and limited labels.
• Probe-based confocal laser endomicroscopy (pCLE) generates images requiring advanced analysis for accurate breast tissue classification.

Purpose of the Study:

• To develop a novel framework for representation learning in medical imaging.
• To improve breast tissue classification accuracy using pCLE images, especially with limited labeled data.
• To align features between pCLE and histology domains for enhanced analysis.

Main Methods:

• A stochastic routing normalization and neighborhood embedding framework was proposed.
• A domain-specific normalization module with stochastic activation was developed for low-level and mid-level feature alignment.
• Latent centers from the histology domain were used as templates for high-level feature matching.

Main Results:

• The proposed method demonstrated superior performance in image classification accuracy on a clinical database of 700 pCLE mosaics.
• Effective domain alignment was achieved between pCLE and histology data.
• The framework successfully learned discriminative features even with limited training samples.

Conclusions:

• The developed framework significantly enhances representation learning for medical image analysis.
• The method offers a promising approach for computer-aided diagnosis in breast tissue classification using pCLE.
• The proposed domain alignment strategy improves classification accuracy with limited data.