Improving Low-contrast Detectability and Noise Texture Pattern for Computed Tomography Using Iterative Reconstruction Accelerated with Machine Learning Method: A Phantom Study

A novel method for developing contrast-detail curves from clinical patient images based on statistical low-contrast detectability

Purpose. To develop a method to extract statistical low-contrast detectability (LCD) and contrast-detail (C-D) curves from clinical patient images.Method. We used the region of air surrounding the patient as an alternative for a homogeneous region within a patient. A simple graphical user interface (GUI) was created to set the initial configuration for region of interest (ROI), ROI size, and minimum detectable contrast (MDC). The process was started by segmenting the air surrounding the patient with a threshold between -980 HU (Hounsfield units) and -1024 HU to get an air mask. The mask was trimmed using the patient center coordinates to avoid distortion from the patient table. It was used to automatically place square ROIs of a predetermined size. The mean pixel values in HU within each ROI were calculated, and the standard deviation (SD) from all the means was obtained. The MDC for a particular target size was generated by multiplying the SD by 3.29. A C-D curve was obtained by iterating this process for the other ROI sizes. This method was applied to the homogeneous area from the uniformity module of an ACR CT phantom to find the correlation between the parameters inside and outside the phantom, for 30 thoracic, 26 abdominal, and 23 head images.Results. The phantom images showed a significant linear correlation between the LCDs obtained from outside and inside the phantom, with R2values of 0.67 and 0.99 for variations in tube currents and tube voltages. This indicated that the air region outside the phantom can act as a surrogate for the homogenous region inside the phantom to obtain the LCD and C-D curves.Conclusion. The C-D curves obtained from outside the ACR CT phantom show a strong linear correlation with those from inside the phantom. The proposed method can also be used to extract the LCD from patient images by using the region of air outside as a surrogate for a region inside the patient.

A Deep Convolutional Gated Recurrent Unit for CT Image Reconstruction

Computed tomography (CT) is one of the most important medical imaging technologies in use today. Most commercial CT products use a technique known as the filtered backprojection (FBP) that is fast and can produce decent image quality when an X-ray dose is high. However, the FBP is not good enough on low-dose X-ray CT imaging because the CT image reconstruction problem becomes more stochastic. A more effective reconstruction technique proposed recently and implemented in a limited number of CT commercial products is an iterative reconstruction (IR). The IR technique is based on a Bayesian formulation of the CT image reconstruction problem with an explicit model of the CT scanning, including its stochastic nature, and a prior model that incorporates our knowledge about what a good CT image should look like. However, constructing such prior knowledge is more complicated than it seems. In this article, we propose a novel neural network for CT image reconstruction. The network is based on the IR formulation and constructed with a recurrent neural network (RNN). Specifically, we transform the gated recurrent unit (GRU) into a neural network performing CT image reconstruction. We call it "GRU reconstruction." This neural network conducts concurrent dual-domain learning. Many deep learning (DL)-based methods in medical imaging are single-domain learning, but dual-domain learning performs better because it learns from both the sinogram and the image domain. In addition, we propose backpropagation through stage (BPTS) as a new RNN backpropagation algorithm. It is similar to the backpropagation through time (BPTT) of an RNN; however, it is tailored for iterative optimization. Results from extensive experiments indicate that our proposed method outperforms conventional model-based methods, single-domain DL methods, and state-of-the-art DL techniques in terms of the root mean squared error (RMSE), the peak signal-to-noise ratio (PSNR), and the structure similarity (SSIM) and in terms of visual appearance.

Physical and visual evaluations of CT image quality of large low-contrast objects with visual model-based iterative reconstruction technique: a phantom study

We aimed to verify whether the image quality of large low-contrast objects can be improved using visual model-based iterative reconstruction (VMR) while maintaining the visibility of conventional filtered back projection (FBP) and reducing radiation dose through physical and visual evaluation. A 64-row multi-slice CT system with SCENARIA View (FUJIFILM healthcare Corp. Tokyo, Japan) was used. The noise power spectrum (NPS), task-based transfer function (TTF), and signal-to-noise ratio (SNR) were physically evaluated. A low contrast object as a substitute for a liver mass was visually evaluated. In the noise measurement, STD1 showed an 18% lower noise compared to FBP. STR4 was able to reduce noise by 58% compared to FBP. The NPS of VMR was similar to those of FBP from low to high spatial frequency. The NPS of VMR reconstructions showed a similar variation with frequency as FBP reconstructions. STD1 showed the highest 10% TTF, and higher 10% TTF was observed with lower VMR level. The SNR of VMR was close to that of FBP, and higher SNR was observed with higher VMR level. In the results of the visual evaluation, there was no significant difference in visual evaluation between STD1 and FBP (p = 0.99) and between STD2 and FBP (p = 0.56). We found that the NPS of VMR images was similar to that of FBP images, and it can reduce noise and radiation dose by 25% and 50%, respectively, without decreasing the visual image quality compared to FBP.

TextureWGAN: texture preserving WGAN with multitask regularizer for computed tomography inverse problems

Purpose

This paper presents a deep learning (DL) based method called TextureWGAN. It is designed to preserve image texture while maintaining high pixel fidelity for computed tomography (CT) inverse problems. Over-smoothed images by postprocessing algorithms have been a well-known problem in the medical imaging industry. Therefore, our method tries to solve the over-smoothing problem without compromising pixel fidelity.

Approach

The TextureWGAN extends from Wasserstein GAN (WGAN). The WGAN can create an image that looks like a genuine image. This aspect of the WGAN helps preserve image texture. However, an output image from the WGAN is not correlated to the corresponding ground truth image. To solve this problem, we introduce the multitask regularizer (MTR) to the WGAN framework to make a generated image highly correlated to the corresponding ground truth image so that the TextureWGAN can achieve high-level pixel fidelity. The MTR is capable of using multiple objective functions. In this research, we adopt a mean squared error (MSE) loss to maintain pixel fidelity. We also use a perception loss to improve the look and feel of result images. Furthermore, the regularization parameters in the MTR are trained along with generator network weights to maximize the performance of the TextureWGAN generator.

Results

The proposed method was evaluated in CT image reconstruction applications in addition to super-resolution and image-denoising applications. We conducted extensive qualitative and quantitative evaluations. We used PSNR and SSIM for pixel fidelity analysis and the first-order and the second-order statistical texture analysis for image texture. The results show that the TextureWGAN is more effective in preserving image texture compared with other well-known methods such as the conventional CNN and nonlocal mean filter (NLM). In addition, we demonstrate that TextureWGAN can achieve competitive pixel fidelity performance compared with CNN and NLM. The CNN with MSE loss can attain high-level pixel fidelity, but it often damages image texture.

Conclusions

TextureWGAN can preserve image texture while maintaining pixel fidelity. The MTR is not only helpful to stabilize the TextureWGAN's generator training but also maximizes the generator performance.

Deep Learning-Based Image Reconstruction for CT Angiography of the Aorta

To evaluate the impact of a novel, deep-learning-based image reconstruction (DLIR) algorithm on image quality in CT angiography of the aorta, we retrospectively analyzed 51 consecutive patients who underwent ECG-gated chest CT angiography and non-gated acquisition for the abdomen on a 256-dectector-row CT. Images were reconstructed with adaptive statistical iterative reconstruction (ASIR-V) and DLIR. Intravascular image noise, the signal-to-noise ratio (SNR) and the contrast-to-noise ratio (CNR) were quantified for the ascending aorta, the descending thoracic aorta, the abdominal aorta and the iliac arteries. Two readers scored subjective image quality on a five-point scale. Compared to ASIR-V, DLIR reduced the median image noise by 51–54% for the ascending aorta and the descending thoracic aorta. Correspondingly, median CNR roughly doubled for the ascending aorta and descending thoracic aorta. There was a 38% reduction in image noise for the abdominal aorta and the iliac arteries, with a corresponding improvement in CNR. Median subjective image quality improved from good to excellent at all anatomical levels. In CT angiography of the aorta, DLIR substantially improved objective and subjective image quality beyond what can be achieved by state-of-the-art iterative reconstruction. This can pave the way for further radiation or contrast dose reductions.