DeepDeblur3D: A 3D U-Net for Denoising and Deblurring Micro-CT Data

Micro-computed tomography (micro-CT) is widely used for non-destructive three-dimensional imaging but is fundamentally constrained by the trade-off between spatial resolution, signal-to-noise ratio, and acquisition time. Post-processing can partially mitigate these limitations; however, classical approaches impose a trade-off between denoising and detail preservation. We present DeepDeblur3D, a standard 3D U-Net trained on 467 heterogeneous micro-CT volumes acquired across multiple scanner platforms and sample types. An inference-time control mechanism allows frequency-selective tuning of denoising and sharpening without retraining. DeepDeblur3D consistently outperforms classical post-processing methods in sharpness, contrast-to-noise ratio, and perceptual similarity, and allows for reliable super-resolution quantitative analysis. The model is released as an open-source Napari plugin, enabling interactive use in routine micro-CT workflows without deep learning expertise.