Three-dimensional imaging is central to modern biomedical research, enabling detailed reconstruction of microanatomical structures and spatial cell relationships. However, the fidelity of these reconstructions often suffers from missing slides, tissue damage, and limited z-resolution, issues compounded by the time and cost associated with high-resolution volumetric imaging.
In a recent study, the authors present InterpolAI, an AI-powered interpolation framework that addresses these limitations by generating synthetic intermediate images between pairs of authentic images in a stack. Built on an optical flow–based architecture adapted for large image motion, InterpolAI restores continuity in 3D datasets, preserves microanatomical detail, and repairs artifacts more effectively than both linear interpolation and the state-of-the-art XVFI approach. Quantitative assessment using Haralick texture features confirms superior preservation of image contrast, luminance, and structural integrity.
The platform shows particular promise in light-sheet microscopy, where it reduces the number of required z-steps, thus significantly decreasing acquisition time for large volumes. This capability could enable faster, more cost-effective imaging without sacrificing resolution.
Limitations remain, InterpolAI requires well-aligned inputs, struggles with large misalignments or rotations, propagates damage when provided with damaged input pairs, and cannot infer rare events absent from both source images. Despite these constraints, the method demonstrates robustness across multiple imaging modalities, species, staining techniques, and resolutions.
By filling gaps and correcting imperfections in biomedical image stacks, InterpolAI not only improves 3D reconstructions but also enhances quantitative downstream analyses such as cellular composition, tissue branching, and topographic mapping. The authors suggest that future developments could extend the approach to spatial omics integration, creating richer multimodal tissue atlases and enabling more comprehensive biological insights.