Sharing massive biomedical data at magnitudes lower bandwidth using implicit neural function.

Area of Science:

• Biomedical data science
• Machine learning for medical imaging
• Data compression algorithms

Background:

• Efficient storage and sharing of large biomedical datasets are crucial for interdisciplinary research.
• Existing compression methods for natural images/videos are inadequate for biomedical data.
• Deep learning methods require extensive data and lack generalizability for biomedical applications.

Purpose of the Study:

• To introduce Biomedical data compRession with Implicit nEural Function (BRIEF) for efficient biomedical data handling.
• To demonstrate BRIEF's capability in achieving high compression ratios with superior fidelity.
• To enable reliable downstream tasks and low-bandwidth data sharing in the biomedical field.

Main Methods:

• Representing biomedical data using compact, data-specific neural networks (implicit neural functions).
• Developing BRIEF to leverage the representational power of neural networks for compression.
• Evaluating compression performance and fidelity across diverse biomedical data types.

Main Results:

• BRIEF achieves 2-3 orders of magnitude compression on diverse biomedical data.
• Achieved significantly higher fidelity compared to existing compression techniques.
• Demonstrated consistent performance across data volumes and supported spatially varying fidelity.

Conclusions:

• BRIEF offers a novel and effective solution for compressing massive biomedical data.
• The method enhances data accessibility and facilitates collaboration through low-bandwidth sharing.
• BRIEF's performance and features promote advancements in the biomedical field.