Densely-Connected Multi-Magnification Hashing for Histopathological Image Retrieval

Area of Science:

• Medical imaging
• Computer-aided diagnosis
• Digital pathology

Background:

• Content-based medical image retrieval aids diagnosis by finding visually similar images.
• Histopathological image retrieval is crucial for breast cancer diagnosis.

Purpose of the Study:

• To propose a novel framework for histopathological image retrieval in breast cancer diagnosis.
• To generate discriminative binary codes using multi-magnification histopathological images.

Main Methods:

• A densely-connected multi-magnification (DCMMH) framework was developed.
• Low-magnification images were enhanced using accumulated similarity from local patches.
• A densely-connected architecture integrated information from high- and low-magnification image pairs.

Main Results:

• The DCMMH framework effectively generated discriminative binary codes.
• Experiments on the BreakHis dataset showed DCMMH's superior performance.
• DCMMH outperformed existing hashing methods for histopathological image retrieval.

Conclusions:

• The proposed DCMMH framework is effective for content-based medical image retrieval in breast cancer diagnosis.
• Integrating multi-magnification information significantly enhances retrieval accuracy.