A body of studies has proposed to obtain high-quality images from low-dose and noisy Computed Tomography (CT) scans for radiation reduction. However, these studies are designed for population-level data without considering the variation in CT devices and individuals, limiting the current approaches' performance, especially for ultra-low-dose CT imaging. Here, we proposed PIMA-CT, a physical anthropomorphic phantom model integrating an unsupervised learning framework, using a novel deep learning technique called Cyclic Simulation and Denoising (CSD), to address these limitations. We first acquired paired low-dose and standard-dose CT scans of the phantom and then developed two generative neural networks: noise simulator and denoiser. The simulator extracts real low-dose noise and tissue features from two separate image spaces (e.g., low-dose phantom model scans and standard-dose patient scans) into a unified feature space. Meanwhile, the denoiser provides feedback to the simulator on the quality of the generated noise. In this way, the simulator and denoiser cyclically interact to optimize network learning and ease the denoiser to simultaneously remove noise and restore tissue features. We thoroughly evaluate our method for removing both real low-dose noise and Gaussian simulated low-dose noise. The results show that CSD outperforms one of the state-of-the-art denoising algorithms without using any labeled data (actual patients' low-dose CT scans) nor simulated low-dose CT scans. This study may shed light on incorporating physical models in medical imaging, especially for ultra-low level dose CT scans restoration.
more »
« less
Generalizability test of a deep learning-based CT image denoising method
Deep learning (DL) has been increasingly explored in low-dose CT image denoising. DL products have also been submitted to the FDA for premarket clearance. While having the potential to improve image quality over the filtered back projection method (FBP) and produce images quickly, generalizability of DL approaches is a major concern because the performance of a DL network can depend highly on the training data. In this work we take a residual encoder-decoder convolutional neural network (REDCNN)-based CT denoising method as an example. We investigate the effect of the scan parameters associated with the training data on the performance of this DL-based CT denoising method and identifies the scan parameters that may significantly impact its performance generalizability. This abstract particularly examines these three parameters: reconstruction kernel, dose level and slice thickness. Our preliminary results indicate that the DL network may not generalize well between FBP reconstruction kernels, but is insensitive to slice thickness for slice-wise denoising. The results also suggest that training with mixed dose levels improves denoising performance.
more »
« less
- Award ID(s):
- 1838179
- NSF-PAR ID:
- 10309183
- Date Published:
- Journal Name:
- The 6th International Conference on Image Formation in X-Ray Computed Tomography
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
In this research, both image denoising and kidney segmentation tasks are addressed jointly via one multitask deep convolutional network. This multitasking scheme yields better results for both tasks compared to separate single-task methods. Also, to the best of our knowledge, this is a first time attempt at addressing these joint tasks in low-dose CT scans (LDCT). This new network is a conditional generative adversarial network (C-GAN) and is an extension of the image-to-image translation network. To investigate the generalized nature of the network, two other conventional single task networks are also exploited, including the well-known 2D UNet method for segmentation and the more recently proposed method WGAN for LDCT denoising. Implementation results proved that the proposed method outperforms UNet and WGAN for both tasks.more » « less
-
more » « less
Purpose To develop a physics‐guided deep learning (PG‐DL) reconstruction strategy based on a signal intensity informed multi‐coil (SIIM) encoding operator for highly‐accelerated simultaneous multislice (SMS) myocardial perfusion cardiac MRI (CMR).
Methods First‐pass perfusion CMR acquires highly‐accelerated images with dynamically varying signal intensity/SNR following the administration of a gadolinium‐based contrast agent. Thus, using PG‐DL reconstruction with a conventional multi‐coil encoding operator leads to analogous signal intensity variations across different time‐frames at the network output, creating difficulties in generalization for varying SNR levels. We propose to use a SIIM encoding operator to capture the signal intensity/SNR variations across time‐frames in a reformulated encoding operator. This leads to a more uniform/flat contrast at the output of the PG‐DL network, facilitating generalizability across time‐frames. PG‐DL reconstruction with the proposed SIIM encoding operator is compared to PG‐DL with conventional encoding operator, split slice‐GRAPPA, locally low‐rank (LLR) regularized reconstruction, low‐rank plus sparse (L + S) reconstruction, and regularized ROCK‐SPIRiT.
Results Results on highly accelerated free‐breathing first pass myocardial perfusion CMR at three‐fold SMS and four‐fold in‐plane acceleration show that the proposed method improves upon the reconstruction methods use for comparison. Substantial noise reduction is achieved compared to split slice‐GRAPPA, and aliasing artifacts reduction compared to LLR regularized reconstruction, L + S reconstruction and PG‐DL with conventional encoding. Furthermore, a qualitative reader study indicated that proposed method outperformed all methods.
Conclusion PG‐DL reconstruction with the proposed SIIM encoding operator improves generalization across different time‐frames /SNRs in highly accelerated perfusion CMR.
-
more » « less
This work concerns a fluorescence optical projection tomography system for low scattering tissue, like lymph nodes, with angular-domain rejection of highly scattered photons. In this regime, filtered backprojection (FBP) image reconstruction has been shown to provide reasonable quality images, yet here a comparison of image quality between images obtained by FBP and iterative image reconstruction with a Monte Carlo generated system matrix, demonstrate measurable improvements with the iterative method. Through simulated and experimental phantoms, iterative algorithms consistently outperformed FBP in terms of contrast and spatial resolution. Moreover, when projection number was reduced, in order to reduce total imaging time, iterative reconstruction suppressed artifacts that hampered the performance of FBP reconstruction (structural similarity of the reconstructed images with “truth” was improved from 0.15 ± 1.2 × 10−3to 0.66 ± 0.02); and although the system matrix was generated for homogenous optical properties, when heterogeneity (62.98 cm-1variance in
µs ) was introduced to simulated phantoms, the results were still comparable (structural similarity homo: 0.67 ± 0.02 vs hetero: 0.66 ± 0.02). -
Currently deployed election systems that scan and process hand-marked ballots are not sophisticated enough to handle marks insufficiently filled in (e.g., partially filled-in), improper marks (e.g., using check marks or crosses instead of filling in bubbles), or marks outside of bubbles, other than setting a threshold to detect whether the pixels inside bubbles are dark and dense enough to be counted as a vote. The current works along this line are still largely limited by their degree of automation and require substantial manpower for annotation and adjudication. In this study, we propose a highly automated deep learning (DL) mark segmentation model-based ballot tabulation assistant able to accurately identify legitimate ballot marks. For comparison purposes, a highly customized traditional computer vision (T-CV) mark segmentation-based method has also been developed to compare with the DL-based tabulator, with a detailed discussion included. Our experiments conducted on two real election datasets achieved the highest accuracy of 99.984% on ballot tabulation. In order to further enhance our DL model’s capability of detecting the marks that are underrepresented in training datasets, e.g., insufficiently or improperly filled marks, we propose a Siamese network architecture that enables our DL model to exploit the contrasting features between a handmarked ballot image and its corresponding blank template image to detect marks. Without the need for extra data collection, by incorporating this novel network architecture, our DL modelbased tabulation method not only achieved a higher accuracy score but also substantially reduced the overall false negative rate.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
