Deep Bayesian Hashing With Center Prior for Multi-Modal Neuroimage Retrieval

Area of Science:

• Medical Imaging
• Computer Science
• Machine Learning

Background:

• Multi-modal neuroimage retrieval aids clinical decision-making by linking images to patient records.
• Existing methods struggle with neuroimages due to small inter-class variation and large inter-modal discrepancy.

Purpose of the Study:

• To develop a robust framework for retrieving information from multi-modal neuroimage databases.
• To address the limitations of current retrieval methods in handling neuroimaging data.

Main Methods:

• Proposed a deep Bayesian hash learning framework (CenterHash) to map multi-modal data into a shared Hamming space.
• Learned discriminative hash codes from imbalanced neuroimages by creating common center representations.
• Utilized a weighted contrastive likelihood loss function for effective hash learning.

Main Results:

• CenterHash successfully generates effective hash codes for multi-modal neuroimages.
• The method demonstrates state-of-the-art performance in cross-modal retrieval tasks.
• Empirical evidence validates the framework's efficacy on three distinct multi-modal neuroimage datasets.

Conclusions:

• CenterHash provides an effective solution for multi-modal neuroimage retrieval.
• The framework enhances the accuracy and efficiency of accessing clinical case information.
• This approach offers significant potential for improving diagnostic and treatment planning processes.