Rectified flows show promise for CT denoising without clean targets

Researchers have explored the use of unconditional rectified flows as a method for denoising low-dose CT scans without requiring clean training targets. Their findings suggest that a single step of the flow is optimal for achieving the minimum mean squared error, and further integration steps can actually increase error. Experiments demonstrated that this one-step approach performs comparably to direct regression on Noise2Noise pairs, outperforming additional Euler steps. While supervised Convolutional Neural Networks (CNNs) achieved significant gains, the label-free performance was highly dependent on how training pairs accounted for correlated CT noise, with fidelity gains not always translating to improved lesion detectability. AI

IMPACT This research could lead to improved medical imaging techniques by enabling more effective denoising of low-dose CT scans without the need for extensive clean data.

RANK_REASON The cluster contains an academic paper detailing a new research methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• CNNS
• CTDenoiser
• Debesh Jha
• Noise2Noise
• peak signal-to-noise ratio
• Structural Similarity Index Measure

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →