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

2025-11-28

AI Summary

Medical image denoising is commonly hindered by the scarcity of annotated clean data and challenges in cross-modality adaptation, limiting the clinical applicability of existing supervised methods. To address this, we propose a training-free, cross-modal generalizable denoising framework that uniquely integrates implicit neural network priors with explicit spectral–spatial dual-domain priors: spectral-domain constraints ensure reconstruction fidelity, while spatial-domain structural regularization preserves anatomical integrity; these are jointly optimized in an end-to-end unsupervised manner. Our method requires neither modality-specific fine-tuning nor ground-truth clean labels. It achieves superior noise suppression and anatomical fidelity across diverse modalities—including CT, MRI, and ultrasound—and various noise types. Moreover, it significantly enhances the robustness and accuracy of downstream tasks such as segmentation and reconstruction.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Security and Privacy: Data transparency and provenance
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

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.

Problem

• Unsupervised denoising for medical imaging without annotated data
• Addressing heterogeneous modalities with limited ground-truth availability
• Preserving frequency characteristics and anatomical structure during reconstruction

Innovation

• Unsupervised denoising via dual-domain prior integration
• Combining implicit architectural and explicit spectral-spatial priors
• No external training data or modality-specific tuning required