In this work, we present a novel self-supervised method for Low Dose Computed
Tomography (LDCT) reconstruction. Reducing the radiation dose to patients
during a CT scan is a crucial challenge since the quality of the reconstruction
highly degrades because of low photons or limited measurements. Supervised deep
learning methods have shown the ability to remove noise in images but require
accurate ground truth which can be obtained only by performing additional
high-radiation CT scans. Therefore, we propose a novel self-supervised
framework for LDCT, in which ground truth is not required for training the
convolutional neural network (CNN). Based on the Noise2Inverse (N2I) method, we
enforce in the training loss the equivariant property of rotation
transformation, which is induced by the CT imaging system, to improve the
quality of the CT image in a lower dose. Numerical and experimental results
show that the reconstruction accuracy of N2I with sparse views is degrading
while the proposed rotational augmented Noise2Inverse (RAN2I) method keeps
better image quality over a different range of sampling angles. Finally, the
quantitative results demonstrate that RAN2I achieves higher image quality
compared to N2I, and experimental results of RAN2I on real projection data show
comparable performance to supervised learning.

本研究提出一种新颖的自监督方法，用于低剂量计算机断层扫描（LDCT）重构，结果表明该方法在降低剂量的情况下能够在 CT 图像质量方面取得更好的效果。

低剂量计算机断层扫描重建的旋转增益噪声 2 反问题

Rotational Augmented Noise2Inverse for Low-dose Computed Tomography  Reconstruction

Recovering a high-quality image from noisy indirect measurements is an
important problem with many applications. For such inverse problems, supervised
deep convolutional neural network (CNN)-based denoising methods have shown
strong results, but the success of these supervised methods critically depends
on the availability of a high-quality training dataset of similar measurements.
For image denoising, methods are available that enable training without a
separate training dataset by assuming that the noise in two different pixels is
uncorrelated. However, this assumption does not hold for inverse problems,
resulting in artifacts in the denoised images produced by existing methods.
Here, we propose Noise2Inverse, a deep CNN-based denoising method for linear
image reconstruction algorithms that does not require any additional clean or
noisy data. Training a CNN-based denoiser is enabled by exploiting the noise
model to compute multiple statistically independent reconstructions. We develop
a theoretical framework which shows that such training indeed obtains a
denoising CNN, assuming the measured noise is element-wise independent and
zero-mean. On simulated CT datasets, Noise2Inverse demonstrates an improvement
in peak signal-to-noise ratio and structural similarity index compared to
state-of-the-art image denoising methods and conventional reconstruction
methods, such as Total-Variation Minimization. We also demonstrate that the
method is able to significantly reduce noise in challenging real-world
experimental datasets.

利用深度卷积神经网络（CNN）开发了一种图像去噪方法：Noise2Inverse，可以应用于线性图像重建算法，无需任何额外的干净或噪声数据，通过利用噪声模型计算多个统计独立的重建来实现训练，该方法在模拟 CT 数据集和现实实验数据集中均显示出优于现有状态下的图像去噪方法和传统重建方法，能够显著降低噪声。