Loading [a11y]/accessibility-menu.js
FISTA-NET: Deep Algorithm Unrolling for Cerenkov luminescence tomography | IEEE Conference Publication | IEEE Xplore

FISTA-NET: Deep Algorithm Unrolling for Cerenkov luminescence tomography

Publisher: IEEE

Abstract:

Cerenkov luminescence tomography (CLT) is a highly sensitive and promising imaging technique that can be used to reconstruct the three-dimensional distribution of radioac...View more

Abstract:

Cerenkov luminescence tomography (CLT) is a highly sensitive and promising imaging technique that can be used to reconstruct the three-dimensional distribution of radioactive probes in living animals. However, the accuracy of CLT reconstruction is limited by the simplified radiative transfer equation and ill-conditioned inverse problem. To address this issue, we propose a model-based deep learning network that combines the neural network with a model-based approach to enhance the performance of CLT reconstruction. The Fast Iterative Shrinkage Thresholding Algorithm (FISTA), a traditional model-based approach, is expanded into a deep network (known as FISTA-NET). Each layer in the network represents an iteration of the algorithm steps, and connecting these layers can form a deep neural network. In addition, different from the traditional FISTA, the key parameters in FISTA, such as gradient step size and threshold value, can be learned through training data without manual production. To evaluate the performance of FISTA-NET, numerical simulation experiments were conducted, which demonstrate its excellent positioning and shape recovery abilities.Clinical Relevance—This indicates that FISTA-NET strategy can significantly improve the quality of CLT reconstruction, which is further beneficial to the assessment of disease activity and treatment effect based on CLT.
Date of Conference: 24-27 July 2023
Date Added to IEEE Xplore: 11 December 2023
ISBN Information:

ISSN Information:

PubMed ID: 38083164
Publisher: IEEE
Conference Location: Sydney, Australia

Funding Agency:


References

References is not available for this document.