Patch-wise Deep Metric Learning for Unsupervised Low-Dose CT Denoising
This addresses CT number distortions that harm diagnostic performance in medical imaging, representing a domain-specific incremental improvement.
The paper tackled the problem of CT number distortions in unsupervised low-dose CT denoising by proposing a patch-wise deep metric learning approach, which achieved high-quality denoised images without CT number shift as confirmed by experimental results.
The acquisition conditions for low-dose and high-dose CT images are usually different, so that the shifts in the CT numbers often occur. Accordingly, unsupervised deep learning-based approaches, which learn the target image distribution, often introduce CT number distortions and result in detrimental effects in diagnostic performance. To address this, here we propose a novel unsupervised learning approach for lowdose CT reconstruction using patch-wise deep metric learning. The key idea is to learn embedding space by pulling the positive pairs of image patches which shares the same anatomical structure, and pushing the negative pairs which have same noise level each other. Thereby, the network is trained to suppress the noise level, while retaining the original global CT number distributions even after the image translation. Experimental results confirm that our deep metric learning plays a critical role in producing high quality denoised images without CT number shift.