Unsupervised Domain Adaptation for Low-Dose Computed Tomography Denoising

Deep neural networks have shown great improvements in low-dose computed tomography (CT) denoising. Early deep learning-based CT denoising algorithms were primarily based on supervised learning. However, learning requires a large number of training samples, which is impractical real-world scenarios. To address this problem, we propose novel unsupervised domain adaptation approach for This proposed framework adapts the network pretrained with paired low- and normal-dose phantom images (source domain) to denoise unlabeled human (target domain). Our modifies action classifier, enabling be adapted target domain. Furthermore, introduce new backpropagation method, call domain-independent weighted backpropagation. By combining these techniques, demonstrate that can properly trained without using clinical clean images. The experimental results showed our method exhibited better performance terms both objective perceptual image qualities when compared current algorithms. represents first-use case context problems, possibility extension other restoration tasks.