Enabling self-supervised learned primal dual with Noise2Inverse

2026-06-25 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, quick

Summary

A new self-supervised reconstruction method, Noise2Inverse Learned Primal-Dual (N2I-LPD), addresses the challenge of X-ray computed tomography (CT) reconstruction in low-dose and sparse-angle scenarios where ground-truth data is scarce. This approach extends the Noise2Inverse framework to the Learned Primal-Dual algorithm, enabling training of an iterative reconstruction operator without ground-truth images. N2I-LPD achieves this by leveraging the statistical independence of noise in distinct measurements, particularly with respect to angular rotation during CT scans. Comparative results show N2I-LPD delivers improved reconstruction quality over classical methods and U-Net-based approaches within the same Noise2Inverse framework, demonstrating the efficacy of combining learned operators with self-supervised training for practical CT imaging.

Key takeaway

For research scientists developing advanced X-ray computed tomography reconstruction algorithms, N2I-LPD offers a robust solution for scenarios lacking ground-truth data. You should consider integrating self-supervised strategies like N2I-LPD into your learned iterative reconstruction pipelines, especially for low-dose and sparse-angle applications. This approach can significantly improve image quality and expand the applicability of learned methods in clinical settings.

Key insights

Self-supervised learning can enable high-quality CT reconstruction without ground-truth data.

Principles

• Exploit noise independence for self-supervision
• Combine learned operators with self-supervised training

Method

Extends Noise2Inverse to the Learned Primal-Dual algorithm, leveraging statistical independence of noise in distinct, angularly rotated CT measurements for training.

In practice

• Reconstruct CT images without ground truth
• Improve low-dose/sparse-angle CT quality

Topics

• X-ray Computed Tomography
• Image Reconstruction
• Self-supervised Learning
• Learned Primal-Dual
• Noise2Inverse
• Inverse Problems

Best for: AI Scientist, Research Scientist

Related on AIssential

• Enabling self-supervised learned primal dual with Noise2Inverse — N2I-LPD enables self-supervised CT reconstruction by extending Noise2Inverse to Learned Primal-Dual, using noise independence for training without ground truth.
• CRIS: Cross-Plane Self-Supervised Isotropic Restoration for Anisotropic Volumetric Imaging Across Modalities — CRIS enables isotropic volumetric image restoration using a self-supervised, cross-plane approach without paired ground truth.
• Fusing Transferred Priors and Physics-based Decomposition for Underwater Image Enhancement — A novel UIE method fuses transferred priors and physics-based decomposition to enhance underwater images without paired labels.
• Plug-and-Play image restoration with Stochastic deNOising REgularization — SNORE improves PnP image restoration by applying denoisers to appropriately noisy images, enhancing performance and theoretical guarantees.
• Deep Learning-based 3D Oral Cavity Reconstruction Using 2D Intraoral Images — Deep learning can reconstruct 3D oral models from minimal 2D images, overcoming cost and discomfort of traditional methods.
• When is 3D Worth It? A Resource-Performance Frontier for CNNs and Transformers in Lung CT — Higher dimensionality in medical imaging models does not guarantee stable performance, especially with class imbalance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.