東華大學圖書館 |

Artificial Intelligence Augmented Medical Imaging Reconstruction in Radiation Therapy /

Record Type:	Electronic resources : Monograph/item
Title/Author:	Artificial Intelligence Augmented Medical Imaging Reconstruction in Radiation Therapy // Di Xu.
Author:	Xu, Di,
Description:	1 electronic resource (167 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 86-12, Section: B.
Contained By:	Dissertations Abstracts International86-12B.
Subject:	Bioengineering. -
Online resource:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31932963
ISBN:	9798280728752

Artificial Intelligence Augmented Medical Imaging Reconstruction in Radiation Therapy /
Xu, Di,

Artificial Intelligence Augmented Medical Imaging Reconstruction in Radiation Therapy /Di Xu. - 1 electronic resource (167 pages)

Source: Dissertations Abstracts International, Volume: 86-12, Section: B.

Background and Purpose: Efficiently acquired and precisely reconstructed imaging are crucial to the success of modern radiation therapy (RT). Computed tomography (CT) and magnetic resonance imaging (MRI) are two common modalities for providing RT treatment planning and delivery guidance/monitoring. In recent decades, artificial intelligence (AI) has emerged as a powerful and widely adopted technique across various fields, valued for its efficiency and convenience enabled by implicit function definition and data-driven feature representation learning. Here, we present a series of AI-driven medical imaging reconstruction frameworks for enhanced radiotherapy, designed to improve CT image reconstruction quality and speed, refine dual-energy CT (DECT) multi-material decomposition (MMD), and significantly accelerate 4D MRI acquisition.Method: Our CT reconstruction framework focuses on ultra-sparse view reconstruction using Neural Radiance Field (NeRF), a powerful technique for reconstructing and rendering 3D scenes from limited views. We developed a novel framework, TomoGRAF, which integrates X-ray transport physics to reconstruct high-quality 3D volumes from ultra-sparse projections without requiring priors. TomoGRAF models CT imaging geometry, simulates the X-ray casting and tracing process, and penalizes the difference between simulated and ground-truth CT sub-volumes during training. Next, we introduce a sinogram-domain material decomposition framework designed for breast tissue differentiation. This non-recursive approach operates directly on raw projection data. Our rFast-MMDNet employs a two-stage algorithm: (1) SinoNet, which performs dual-energy projection decomposition on tissue sinograms, and (2) FBP-DenoiseNet, which facilitates domain adaptation and image post-processing.Finally, we developed three distinct 4D MRI acceleration pipelines leveraging Transformers, generative adversarial networks (GANs), and diffusion networks. For the Transformer-based approach, we propose Reconstruction Swin Transformer (RST), which builds upon the Video Swin Transformer architecture with a novel reconstruction head to restore pixel-wise intensity. To enhance efficiency, we introduce SADXNet, a convolutional network that rapidly initializes 2D MR frames before RST learning, significantly reducing model complexity, GPU demand, and training time. For the GAN-based pipeline, we present Reconstruct Paired Conditional Generative Adversarial Network (Re-Con-GAN), which accelerates 4D MRI reconstruction while preserving image quality. Re-Con-GAN is trained using paired input-output data, exploring ResNet9, UNet, and RST as potential generators, with PatchGAN as the discriminator. For diffusion-based reconstruction, we introduce the Chained Iterative Reconstruction Network (CIRNet), designed for efficient sparse-sampling reconstruction while maintaining clinically deployable quality. CIRNet employs a denoising diffusion probabilistic framework, conditioning image reconstruction through a stochastic iterative denoising process. During training, a forward Markovian diffusion process progressively adds Gaussian noise to the densely sampled ground truth, while CIRNet learns to reverse this process iteratively. At inference, CIRNet runs the reverse process to recover signals from noise, conditioned on the undersampled input.Result: In the task of TomoGRAF, we evaluated the performance of TomoGRAF on an unseen dataset of distinct imaging characteristics from the training data and demonstrated a vast leap in performance compared with state-of-the-art deep learning and NeRF methods. TomoGRAF provides the first generalizable solution for image-guided radiotherapy and interventional radiology applications, where only one or a few X-ray views are available, but 3D volumetric information is desired. For rFast-MMDNet, MMD for breast fibroglandular, adipose tissues, and calcification was performed using the 2022 DL-Spectral-Challenge dataset. Our results show that models trained from the sinogram domain can yield high-fidelity decomposition where rFast-MMDNet is particularly dominant. In our 4D MRI acceleration project series, RST was evaluated on a lung respiratory dataset, demonstrating promising reconstruction performance. Re-Con-GAN and the diffusion-based CIRNet were tested on 4D liver MRI, where Re-Con-GAN achieved sub-second inference while maintaining high-quality reconstruction at 10x acceleration. Meanwhile, CIRNet, though requiring slightly longer inference time (~10 seconds), pushed acceleration beyond 30x while preserving high reconstruction quality.Conclusion: The proposed AI frameworks achieve fast image reconstruction with favorable quality in diverse imaging modality including CT, DECT and MRI.

English

ISBN: 9798280728752Subjects--Topical Terms:

657580
Bioengineering.
Subjects--Index Terms:

Computed tomography

Artificial Intelligence Augmented Medical Imaging Reconstruction in Radiation Therapy /
LDR:06242nmm a22004453i 4500 001 2434553
005 20250912102040.5
006 m o d
007 cr|nu||||||||
008 261011s2025 miu||||||m |||||||eng d
020 $a 9798280728752
035 $a (MiAaPQD)AAI31932963
035 $a AAI31932963
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Xu, Di, $e author. $0 (orcid)0000-0001-5587-5275 $3 3824562
245 1 0 $a Artificial Intelligence Augmented Medical Imaging Reconstruction in Radiation Therapy / $c Di Xu.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2025
300 $a 1 electronic resource (167 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 86-12, Section: B.
500 $a Advisors: Sheng, Ke Committee members: Descovich, Martina; Yang, Yang; Liu, Chunlei.
502 $b Ph.D. $c University of California, San Francisco $d 2025.
520 $a Background and Purpose: Efficiently acquired and precisely reconstructed imaging are crucial to the success of modern radiation therapy (RT). Computed tomography (CT) and magnetic resonance imaging (MRI) are two common modalities for providing RT treatment planning and delivery guidance/monitoring. In recent decades, artificial intelligence (AI) has emerged as a powerful and widely adopted technique across various fields, valued for its efficiency and convenience enabled by implicit function definition and data-driven feature representation learning. Here, we present a series of AI-driven medical imaging reconstruction frameworks for enhanced radiotherapy, designed to improve CT image reconstruction quality and speed, refine dual-energy CT (DECT) multi-material decomposition (MMD), and significantly accelerate 4D MRI acquisition.Method: Our CT reconstruction framework focuses on ultra-sparse view reconstruction using Neural Radiance Field (NeRF), a powerful technique for reconstructing and rendering 3D scenes from limited views. We developed a novel framework, TomoGRAF, which integrates X-ray transport physics to reconstruct high-quality 3D volumes from ultra-sparse projections without requiring priors. TomoGRAF models CT imaging geometry, simulates the X-ray casting and tracing process, and penalizes the difference between simulated and ground-truth CT sub-volumes during training. Next, we introduce a sinogram-domain material decomposition framework designed for breast tissue differentiation. This non-recursive approach operates directly on raw projection data. Our rFast-MMDNet employs a two-stage algorithm: (1) SinoNet, which performs dual-energy projection decomposition on tissue sinograms, and (2) FBP-DenoiseNet, which facilitates domain adaptation and image post-processing.Finally, we developed three distinct 4D MRI acceleration pipelines leveraging Transformers, generative adversarial networks (GANs), and diffusion networks. For the Transformer-based approach, we propose Reconstruction Swin Transformer (RST), which builds upon the Video Swin Transformer architecture with a novel reconstruction head to restore pixel-wise intensity. To enhance efficiency, we introduce SADXNet, a convolutional network that rapidly initializes 2D MR frames before RST learning, significantly reducing model complexity, GPU demand, and training time. For the GAN-based pipeline, we present Reconstruct Paired Conditional Generative Adversarial Network (Re-Con-GAN), which accelerates 4D MRI reconstruction while preserving image quality. Re-Con-GAN is trained using paired input-output data, exploring ResNet9, UNet, and RST as potential generators, with PatchGAN as the discriminator. For diffusion-based reconstruction, we introduce the Chained Iterative Reconstruction Network (CIRNet), designed for efficient sparse-sampling reconstruction while maintaining clinically deployable quality. CIRNet employs a denoising diffusion probabilistic framework, conditioning image reconstruction through a stochastic iterative denoising process. During training, a forward Markovian diffusion process progressively adds Gaussian noise to the densely sampled ground truth, while CIRNet learns to reverse this process iteratively. At inference, CIRNet runs the reverse process to recover signals from noise, conditioned on the undersampled input.Result: In the task of TomoGRAF, we evaluated the performance of TomoGRAF on an unseen dataset of distinct imaging characteristics from the training data and demonstrated a vast leap in performance compared with state-of-the-art deep learning and NeRF methods. TomoGRAF provides the first generalizable solution for image-guided radiotherapy and interventional radiology applications, where only one or a few X-ray views are available, but 3D volumetric information is desired. For rFast-MMDNet, MMD for breast fibroglandular, adipose tissues, and calcification was performed using the 2022 DL-Spectral-Challenge dataset. Our results show that models trained from the sinogram domain can yield high-fidelity decomposition where rFast-MMDNet is particularly dominant. In our 4D MRI acceleration project series, RST was evaluated on a lung respiratory dataset, demonstrating promising reconstruction performance. Re-Con-GAN and the diffusion-based CIRNet were tested on 4D liver MRI, where Re-Con-GAN achieved sub-second inference while maintaining high-quality reconstruction at 10x acceleration. Meanwhile, CIRNet, though requiring slightly longer inference time (~10 seconds), pushed acceleration beyond 30x while preserving high reconstruction quality.Conclusion: The proposed AI frameworks achieve fast image reconstruction with favorable quality in diverse imaging modality including CT, DECT and MRI.
546 $a English
590 $a School code: 0034
650 4 $a Bioengineering. $3 657580
650 4 $a Biomedical engineering. $3 535387
650 4 $a Medical imaging. $3 3172799
653 $a Computed tomography
653 $a Deep learning
653 $a Magnetic resonance imaging
653 $a Radiation therapy
653 $a Reconstruction
690 $a 0202
690 $a 0541
690 $a 0574
690 $a 0800
710 2 $a University of California, San Francisco. $b Bioengineering. $e degree granting institution. $3 3824563
720 1 $a Sheng, Ke $e degree supervisor.
773 0 $t Dissertations Abstracts International $g 86-12B.
790 $a 0034
791 $a Ph.D.
792 $a 2025
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31932963