A deep learning post-processing to enhance the maximum likelihood estimate of three material decomposition in photon counting spectral CT

Eguizabal, Alma; Persson, Mats; Grönberg, Fredrik

doi:10.1117/12.2581044

Cited by 4 publications

(6 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We aim to explore two "physics-aware" deep learning approaches of this type, both based on unrolling a suitable iterative scheme. The first approach considers a neural network architecture given by considering a learned update to a gradient-descent scheme as in [32]. The second approach includes a more explicit use of the physics with a learned update function that also casts the value of the gradient of the likelihood cost used in the model-based approach.…”

Section: Proposed Deep Learning Solutionsmentioning

confidence: 99%

“…In order to keep the test case challenging, and assuming that in medical imaging training data is not abundant (an especially in an emerging technology such as PCCT), we have considered only 200 training samples, which is a relative small number (in opposition to our previous work in [32] and [33]). We use 100 test cases to illustrate and compute the results.…”

Section: Implementation and Evaluationmentioning

confidence: 99%

“…The technique is evaluated for a simulated dataset generated from a realistic model of a silicon-based PCCT system with eight energy bins, incorporating spectral distortion due to non-ideal effects such as Compton scatter. Preliminary investigations of the proposed method have been presented previously in [32] and [33].…”

mentioning

confidence: 99%

See 2 more Smart Citations

Deep Learning for Material Decomposition in Photon-Counting CT

Eguizabal¹,

Öktem²,

Persson³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

Photon-counting CT (PCCT) offers improved diagnostic performance through better spatial and energy resolution, but developing high-quality image reconstruction methods that can deal with these large datasets is challenging. Model-based solutions incorporate models of the physical acquisition in order to reconstruct more accurate images, but are dependent on an accurate forward operator and present difficulties with finding good regularization. Another approach is deep-learning reconstruction, which has shown great promise in CT. However, fully data-driven solutions typically need large amounts of training data and lack interpretability. To combine the benefits of both methods, while minimizing their respective drawbacks, it is desirable to develop reconstruction algorithms that combine both model-based and data-driven approaches. In this work, we present a novel deep-learning solution for material decomposition in PCCT, based on an unrolled/unfolded iterative network. We evaluate two cases: a learned post-processing, which implicitly utilizes model knowledge, and a learned gradient-descent, which has explicit model-based components in the architecture. With our proposed techniques, we solve a challenging PCCT simulation case: three-material decomposition in abdomen imaging with low dose, iodine contrast, and a very small training sample support. In this scenario, our approach outperforms a maximum likelihood estimation, a variational method, as well as a fully-learned network.

show abstract

Section: Proposed Deep Learning Solutionsmentioning

confidence: 99%

Section: Implementation and Evaluationmentioning

confidence: 99%

See 1 more Smart Citation

Deep Learning for Material Decomposition in Photon-Counting CT

Eguizabal¹,

Öktem²,

Persson³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Spectral CT in particular has seen relatively little work. Some works use deep learning for material decomposition, both in imageand projection-space [16], [15], [27].…”

Section: Modelling Physics With Machine Learningmentioning

confidence: 99%

“…Whereas the training phantoms are simulating the cross section of a head and therefore all have a thin outer shell made of bone, this data simulates a human torso, containing inner cavities, more intricate skeletal structures and non-binary concentration values. The phantom has been artificially separated into material channels using the procedure described in [27]. Examples from the data along with simulated measurements are shown in Figure 1.…”

Section: B Datamentioning

confidence: 99%

Photon-Counting CT Reconstruction With a Learned Forward Operator

Ström

Persson

Eguizabal

et al. 2022

IEEE Trans. Comput. Imaging

Self Cite

View full text Add to dashboard Cite

Photon-Counting CT is an emerging imaging technology that promises higher spatial resolution and the possibility for material decomposition in the reconstruction. A major difficulty in Photon-Counting CT is to efficiently model cross-talk between detectors. In this work, we accelerate image reconstruction tasks for Photon-Counting CT by modelling the cross-talk with an appropriately trained deep convolutional neural network. The main result relates to proving convergence when using such a learned cross-talk model in the context of second-order optimisation methods for spectral CT. Another is to evaluate the method through numerical experiments on small-scale CT acquisitions generated using a realistic physics model. Using the reconstruction with a full cross-talk model as ground truth, the learned cross-talk model results in a 20 dB increase in peaksignal-to noise ratio compared to ignoring cross-talk altogether. At the same time, it effectively cuts the computation time of the full cross-talk model in half. Furthermore, the learned crosstalk model generalises well to both unseen data and unseen detector settings. Our results indicate that such a partially learned forward operator is a suitable way of modelling data generation in Photon-Counting CT with a computational benefit that becomes more noticeable for realistic problem sizes.

show abstract