Noise2Stack: Improving Image Restoration by Learning from Volumetric Data

Biomedical images are noisy. The imaging equipment itself has physical limitations, and the consequent experimental trade-offs between signal-to-noise ratio, acquisition speed, depth exacerbate problem. Denoising is, therefore, an essential part of any image processing pipeline, convolutional neural networks currently method choice for this task. One popular approach, Noise2Noise, does not require clean ground truth, instead, uses a second noisy copy as training target. Self-supervised methods, like Noise2Self Noise2Void, learn signal without explicit target, but limited by lack information in single image. Here, we introduce Noise2Stack, extension Noise2Noise to stacks that takes advantage shared spatially neighboring planes. Our experiments on magnetic resonance brain scans multiplane microscopy data show learning only from neighbors stack is sufficient outperform Noise2Void close gap supervised denoising methods. findings point low-cost, high-reward improvement pipelines biomedical images.