An Innovative Attention-based Triplet Deep Hashing Approach to Retrieve Histopathology Images

Seyed Mohammad Alizadeh ¹, Mohammad Sadegh Helfroush ², M Emre Celebi ³

⁰Department of Electrical Engineering, Shiraz University of Technology, Shiraz, Iran. s.alizadeh@sutech.ac.ir.

Journal of Imaging Informatics in Medicine +

|

November 11, 2024

View abstract on PubMed

Summary

Histopathology attention triplet deep hashing (HATDH) improves disease diagnosis by efficiently retrieving images. This novel method uses deep learning with attention mechanisms and triplet loss for superior performance over existing algorithms.

Area Of Science

Digital pathology
Medical image analysis
Computer-assisted diagnosis

Background

Content-based histopathology image retrieval (CBHIR) aids disease diagnosis but can be complex with high-dimensional features.
Hashing techniques reduce dimensionality by mapping features to binary values, with deep learning methods outperforming traditional ones.
Triplet-based deep hashing models are generally more effective than pairwise approaches, and attention mechanisms further enhance retrieval accuracy.

Purpose Of The Study

To introduce an innovative triplet deep hashing strategy incorporating an attention mechanism for histopathology image retrieval.
To enhance the efficiency and accuracy of content-based histopathology image retrieval.
To develop a method that overcomes the complexity and time-consumption associated with high-dimensional features in histopathology image analysis.

Main Methods

Developed Histopathology Attention Triplet Deep Hashing (HATDH), a novel strategy for histopathology image retrieval.
Employed three deep attention-based hashing models with identical architectures and weights to generate binary hash values.
Introduced an improved triplet loss function that considers both triplet and pair inputs for enhanced training and retrieval efficiency.

Main Results

The proposed attention module facilitates more efficient feature extraction within the deep hashing models.
The improved triplet loss function enhances efficiency during both training and retrieval phases.
HATDH demonstrated significantly superior performance compared to state-of-the-art hashing algorithms on the BreakHis and Kather datasets.

Conclusions

HATDH offers a significant advancement in content-based histopathology image retrieval.
The integration of attention mechanisms and an improved triplet loss function leads to enhanced retrieval accuracy and efficiency.
This approach holds promise for improving computer-assisted diagnosis in digital pathology.

Keywords:

Attention mechanism Content-based histopathology image retrieval Hashing algorithms Neural networks Triplet models