DNA-Prior: Unsupervised Denoise Anything via Dual-Domain Prior

TL;DR

DNA-Prior presents a universal unsupervised denoising method for medical imaging by integrating an implicit architectural prior with an explicit spectral-spatial regulariser. Reconstruction is constrained to a CNN-generated image manifold and optimised via a dual-domain loss combining a frequency-domain fidelity term with a total-variation penalty. The method demonstrates robust, modality-agnostic denoising across seven imaging modalities under varying Gaussian noise without ground-truth data, outperforming traditional explicit-prior and purely implicit methods. This yields improved noise suppression while preserving anatomical detail, with potential to streamline clinical imaging pipelines.

Abstract

Medical imaging pipelines critically rely on robust denoising to stabilise downstream tasks such as segmentation and reconstruction. However, many existing denoisers depend on large annotated datasets or supervised learning, which restricts their usability in clinical environments with heterogeneous modalities and limited ground-truth data. To address this limitation, we introduce DNA-Prior, a universal unsupervised denoising framework that reconstructs clean images directly from corrupted observations through a mathematically principled hybrid prior. DNA-Prior integrates (i) an implicit architectural prior, enforced through a deep network parameterisation, with (ii) an explicit spectral-spatial prior composed of a frequency-domain fidelity term and a spatial regularisation functional. This dual-domain formulation yields a well-structured optimisation problem that jointly preserves global frequency characteristics and local anatomical structure, without requiring any external training data or modality-specific tuning. Experiments across multiple modalities show that DNA achieves consistent noise suppression and structural preservation under diverse noise conditions.

Figures (1)

• Figure 1: Visual comparison of denoising results for $\sigma = 50$, showing DNA-Prior versus existing state-of-the-art methods.