Tada-DIP: Input-adaptive Deep Image Prior for One-shot 3D Image Reconstruction

TL;DR

Tada-DIP addresses the challenge of reconstructing 3D images from undersampled measurements by extending Deep Image Prior into a fully 3D framework that couples input-adaptation with denoising regularization. The method uses a 3D U-Net backbone, injects input noise to enforce denoising, and updates the input iteratively to guide optimization, achieving high-quality reconstructions without training data. Empirical results on sparse-view 3D CT show strong performance, outperforming dataless baselines and closely approaching a supervised network trained on full datasets. This suggests Tada-DIP as a robust, training-free approach for large-scale 3D reconstruction tasks and motivates further theoretical and modality-specific extensions.

Abstract

Deep Image Prior (DIP) has recently emerged as a promising one-shot neural-network based image reconstruction method. However, DIP has seen limited application to 3D image reconstruction problems. In this work, we introduce Tada-DIP, a highly effective and fully 3D DIP method for solving 3D inverse problems. By combining input-adaptation and denoising regularization, Tada-DIP produces high-quality 3D reconstructions while avoiding the overfitting phenomenon that is common in DIP. Experiments on sparse-view X-ray computed tomography reconstruction validate the effectiveness of the proposed method, demonstrating that Tada-DIP produces much better reconstructions than training-data-free baselines and achieves reconstruction performance on par with a supervised network trained using a large dataset with fully-sampled volumes.

Figures (8)

• Figure 1: Overview of the setup of the proposed algorithm, Tada-DIP.
• Figure 2: Block diagram illustrating one iteration of the proposed Tada-DIP algorithm.
• Figure 3: 3D visualization (maximum intensity projection) of one $256^3$ test volume reconstructed using the proposed Tada-DIP from 30 views. The softer tissues have been thresholded away. Zoom-ins highlight Tada-DIP's ability to faithfully reconstruct fine details.
• Figure 4: Qualitative comparison of reconstruction methods for 30-view parallel beam CT reconstruction. The visualization shows one axial slice of the 3D reconstruction.
• Figure 5: Visualization of coronal slices for 30-view parallel beam reconstruction (same volume as \ref{['fig:qual_comp_30_axial']}). Note Tada-DIP's ability to accurately reconstruct fine details, whereas the supervised reconstruction appears overly smooth.