Improving Tomographic Reconstruction from Limited Data Using Mixed-Scale Dense Convolutional Neural Networks

Pelt, Daniël M.; Batenburg, Kees Joost; Sethian, James A.

doi:10.3390/jimaging4110128

Cited by 96 publications

(95 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, the network has shown good results for removing noise and other artifacts from tomographic images when a large training set of similar objects scanned at high dose is available. This is described in [34], where the network was applied to tomographic images reconstructed from a dataset of parallel beam projections, rather than cone beam projections.…”

Section: Machine Learningmentioning

confidence: 99%

On-the-Fly Machine Learning for Improving Image Resolution in Tomography

et al. 2019

View full text Add to dashboard Cite

In tomography, the resolution of the reconstructed 3D volume is inherently limited by the pixel resolution of the detector and optical phenomena. Machine learning has demonstrated powerful capabilities for super-resolution in several imaging applications. Such methods typically rely on the availability of high-quality training data for a series of similar objects. In many applications of tomography, existing machine learning methods cannot be used because scanning such a series of similar objects is either impossible or infeasible. In this paper, we propose a novel technique for improving the resolution of tomographic volumes that is based on the assumption that the local structure is similar throughout the object. Therefore, our approach does not require a training set of similar objects. The technique combines a specially designed scanning procedure with a machine learning method for super-resolution imaging. We demonstrate the effectiveness of our approach using both simulated and experimental data. The results show that the proposed method is able to significantly improve resolution of tomographic reconstructions.

show abstract

Section: Machine Learningmentioning

confidence: 99%

On-the-Fly Machine Learning for Improving Image Resolution in Tomography

et al. 2019

View full text Add to dashboard Cite

show abstract

“…We modeled the Bessel beam as a constant line segment for computational simplicity, but models of the actual point spread function could be used for computing the inverse Radon transform. Approaches combining imaging physics and machine learning have been successfully applied in other tomography techniques for improving the reconstruction from sparse, shallow angle projections [13,[16][17][18][19][20]. Our networks were limited by GPU memory and therefore only low resolution images (128 pixel resolution instead of 512) were used.…”

Section: Discussionmentioning

confidence: 99%

“…Volume information is obtained from four independent projections recorded from four different angles using temporally multiplexed, tilted Bessel beams in a single frame scan. For volume reconstruction we combine inverse Radon transforms adapted for Bessel beam scanning with machine learning [13,[16][17][18][19][20]. Machine learning has been shown for example in optical phase imaging to allow high resolution reconstruction from sparse projections at shallow angles similar to the ones used here [18].…”

Section: Introductionmentioning

confidence: 99%

Bessel beam tomography for fast volume imaging

Valle

Seelig

2019

Preprint

View full text Add to dashboard Cite

Light microscopy on dynamic samples, for example neural activity in the brain, requires imaging large volumes at high rates. Here, we develop a tomography approach for scanning fluorescence microscopy which allows recording volume images at frame scan rates. Volumes are imaged by simultaneously recording four independent projections at different angles using temporally multiplexed, tilted Bessel beams. From the resulting projections, volumes are reconstructed using inverse Radon transforms combined with three dimensional convolutional neural networks (U-net). This tomography approach is suitable for experiments requiring fast volume imaging of sparse samples, as for example often encountered when imaging neural activity in the brain.

show abstract

“…TomoGAN outperforms FBPConvNet on PSNR (higher median and smaller variance means more stable performance) but has a worse SSIM score. However, when we examine an arbitrarily chosen denoised image as shown in Figure 18, we see that FBPConvNet introduces low-contrast artifacts as marked in Figure 18(a) (there are visibly more white dots in Figure 18(a) than in Figure 18(d)), likely because: (1) FBPConvNet only used MSE loss and those artifacts were not significant to the MSE, and/or (2) as observed by the FBPConvNet authors [15], FBPConvNet has a high risk of overfitting. But these artifacts are significant to TomoGAN's adversarial loss and perceptual loss.…”

Section: Comparison With Other Solutions On Experimental Datasetsmentioning

confidence: 93%

“…They achieved a 10-fold increase in signal-to-noise ratio, enabling the reliable tracing of brain structures in low-dose datasets. For (iii), Pelt et al [15] trained a mixed-scale dense convolutional neural network [16] in a supervised fashion to learn a mapping from low-dose to normal-dose reconstructions. They achieved impressive results on simulation datasets; but unfortunately the performance of proposed model is not thoroughly evaluated on different experimental datasets.…”

mentioning

confidence: 99%

TomoGAN: low-dose synchrotron x-ray tomography with generative adversarial networks: discussion

et al. 2020

View full text Add to dashboard Cite

Synchrotron-based x-ray tomography is a noninvasive imaging technique that allows for reconstructing the internal structure of materials at high spatial resolutions from tens of micrometers to a few nanometers. In order to resolve sample features at smaller length scales, however, a higher radiation dose is required. Therefore, the limitation on the achievable resolution is set primarily by noise at these length scales. We present TomoGAN, a denoising technique based on generative adversarial networks, for improving the quality of reconstructed images for low-dose imaging conditions. We evaluate our approach in two photon-budget-limited experimental conditions: (1) sufficient number of low-dose projections (based on Nyquist sampling), and (2) insufficient or limited number of high-dose projections. In both cases the angular sampling is assumed to be isotropic, and the photon budget throughout the experiment is fixed based on the maximum allowable radiation dose on the sample. Evaluation with both simulated and experimental datasets shows that our approach can significantly reduce noise in reconstructed images, improving the structural similarity score of simulation and experimental data from 0.18 to 0.9 and from 0.18 to 0.41, respectively. Furthermore, the quality of the reconstructed images with filtered back projection followed by our denoising approach exceeds that of reconstructions with the simultaneous iterative reconstruction technique, showing the computational superiority of our approach.

show abstract

Improving Tomographic Reconstruction from Limited Data Using Mixed-Scale Dense Convolutional Neural Networks

Cited by 96 publications

References 33 publications

On-the-Fly Machine Learning for Improving Image Resolution in Tomography

On-the-Fly Machine Learning for Improving Image Resolution in Tomography

Bessel beam tomography for fast volume imaging

TomoGAN: low-dose synchrotron x-ray tomography with generative adversarial networks: discussion

Contact Info

Product

Resources

About