InterpolAI: deep learning-based optical flow interpolation and restoration of biomedical images for improved 3D...

Area of Science:

• Biological imaging
• Computational biology
• Artificial intelligence

Background:

• Accurate analysis of 3D biological datasets relies on high-quality imaging.
• Image quality is often compromised by missing data, tissue damage, or low resolution.

Purpose of the Study:

• To introduce InterpolAI, a novel method for improving the quality of 3D biological image datasets.
• To address limitations in image quality affecting the accuracy of 3D imaging analyses.

Main Methods:

• InterpolAI leverages an optical flow-based artificial intelligence (AI) model, specifically frame interpolation for large image motion.
• The method generates synthetic images between authentic image pairs in a 3D stack.

Main Results:

• InterpolAI outperforms linear interpolation and the XVFI method in preserving microanatomical features, cell counts, contrast, variance, and luminance.
• The AI model effectively repairs tissue damage and reduces stitching artifacts.
• Validation across diverse imaging modalities, species, staining techniques, and resolutions confirmed InterpolAI's robustness.

Conclusions:

• InterpolAI significantly enhances the resolution, throughput, and quality of 3D biological image datasets.
• The AI-driven approach holds substantial potential for advancing 3D imaging analysis and discovery.
• This method offers a powerful tool for overcoming common challenges in biological imaging.