Transformers in medical imaging: A survey

Medical Image Analysis, Journal Year: 2023, Volume and Issue: 88, P. 102802 - 102802

Published: April 5, 2023

Language: Английский

---

Unsupervised Medical Image Translation With Adversarial Diffusion Models

IEEE Transactions on Medical Imaging, Journal Year: 2023, Volume and Issue: 42(12), P. 3524 - 3539

Published: June 28, 2023

Imputation of missing images via source-to-target modality translation can improve diversity in medical imaging protocols. A pervasive approach for synthesizing target involves one-shot mapping through generative adversarial networks (GAN). Yet, GAN models that implicitly characterize the image distribution suffer from limited sample fidelity. Here, we propose a novel method based on diffusion modeling, SynDiff, improved performance translation. To capture direct correlate distribution, SynDiff leverages conditional process progressively maps noise and source onto image. For fast accurate sampling during inference, large steps are taken with projections reverse direction. enable training unpaired datasets, cycle-consistent architecture is devised coupled diffusive non-diffusive modules bilaterally translate between two modalities. Extensive assessments reported utility against competing multi-contrast MRI MRI-CT Our demonstrations indicate offers quantitatively qualitatively superior baselines.

Language: Английский

---

Adaptive diffusion priors for accelerated MRI reconstruction

Medical Image Analysis, Journal Year: 2023, Volume and Issue: 88, P. 102872 - 102872

Published: June 20, 2023

Language: Английский

---

BolT: Fused window transformers for fMRI time series analysis

Medical Image Analysis, Journal Year: 2023, Volume and Issue: 88, P. 102841 - 102841

Published: May 18, 2023

Language: Английский

---

Recent developments of the reconstruction in magnetic particle imaging

Visual Computing for Industry Biomedicine and Art, Journal Year: 2022, Volume and Issue: 5(1)

Published: Oct. 1, 2022

Abstract Magnetic particle imaging (MPI) is an emerging molecular technique with high sensitivity and temporal-spatial resolution. Image reconstruction important research topic in MPI, which converts induced voltage signal into the image of superparamagnetic iron oxide particles concentration distribution. MPI primarily involves system matrix- x-space-based methods. In this review, we provide a detailed overview status future trends these two addition, review application deep learning methods current open sources MPI. Finally, opinions on are presented. We hope promotes use clinical applications.

Language: Английский

---

Deep learning based MRI reconstruction with transformer

Computer Methods and Programs in Biomedicine, Journal Year: 2023, Volume and Issue: 233, P. 107452 - 107452

Published: March 1, 2023

Language: Английский

---

One model to unite them all: Personalized federated learning of multi-contrast MRI synthesis

Medical Image Analysis, Journal Year: 2024, Volume and Issue: 94, P. 103121 - 103121

Published: Feb. 23, 2024

Language: Английский

---

Progressive Pretraining Network for 3D System Matrix Calibration in Magnetic Particle Imaging

IEEE Transactions on Medical Imaging, Journal Year: 2023, Volume and Issue: 42(12), P. 3639 - 3650

Published: July 20, 2023

Magnetic particle imaging (MPI) is an emerging technique for determining magnetic nanoparticle distributions in biological tissues. Although system-matrix (SM)-based image reconstruction offers higher quality than the X-space-based approach, SM calibration measurement time-consuming. Additionally, should be recalibrated if tracer's characteristics or field environment change, and repeated further increase required labor time. Therefore, fast essential MPI. Existing methods commonly treat each row of as independent others, but rows are inherently related through coil channel frequency index. As these two elements can regarded additional multimodal information, we leverage transformer architecture with a self-attention mechanism to encode them. has shown superiority fusion learning across several fields, its high complexity may lead overfitting when labeled data scarce. Compared (i.e., full size), low-resolution easily obtained, fully using such alleviate overfitting. Accordingly, propose pseudo-label-based progressive pretraining strategy unlabeled data. Our method outperforms existing on public real-world OpenMPI dataset simulation dataset. Moreover, our improves resolution in-house MPI scanners without requiring full-size measurements. Ablation studies confirm contributions modeling inter-row relations proposed strategy.

Language: Английский

---

DEQ-MPI: A Deep Equilibrium Reconstruction With Learned Consistency for Magnetic Particle Imaging

IEEE Transactions on Medical Imaging, Journal Year: 2023, Volume and Issue: 43(1), P. 321 - 334

Published: Aug. 1, 2023

Magnetic particle imaging (MPI) offers unparalleled contrast and resolution for tracing magnetic nanoparticles. A common procedure calibrates a system matrix (SM) that is used to reconstruct data from subsequent scans. The ill-posed reconstruction problem can be solved by simultaneously enforcing consistency based on the SM regularizing solution an image prior. Traditional hand-crafted priors cannot capture complex attributes of MPI images, whereas recent methods learned suffer extensive inference times or limited generalization performance. Here, we introduce novel physics-driven method deep equilibrium model with (DEQ-MPI). DEQ-MPI reconstructs images augmenting neural networks into iterative optimization, as inspired unrolling in learning. Yet, conventional are computationally restricted few iterations resulting non-convergent solutions, they use measures yield suboptimal distribution. instead trains implicit mapping maximize quality convergent solution, it incorporates measure better account Demonstrations simulated experimental indicate achieves superior competitive time state-of-the-art methods.

Language: Английский

---

Hierarchical Perception Adversarial Learning Framework for Compressed Sensing MRI

IEEE Transactions on Medical Imaging, Journal Year: 2023, Volume and Issue: 42(6), P. 1859 - 1874

Published: Jan. 30, 2023

The long acquisition time has limited the accessibility of magnetic resonance imaging (MRI) because it leads to patient discomfort and motion artifacts. Although several MRI techniques have been proposed reduce time, compressed sensing in (CS-MRI) enables fast without compromising SNR resolution. However, existing CS-MRI methods suffer from challenge aliasing This results noise-like textures missing fine details, thus leading unsatisfactory reconstruction performance. To tackle this challenge, we propose a hierarchical perception adversarial learning framework (HP-ALF). HP-ALF can perceive image information mechanism: image-level patch-level perception. former visual difference entire image, achieve artifact removal. latter regions recover details. Specifically, achieves mechanism by utilizing multilevel perspective discrimination. discrimination provide two perspectives (overall regional) for learning. It also utilizes global local coherent discriminator structure generator during training. In addition, contains context-aware block effectively exploit slice between individual images better experiments validated on three datasets demonstrate effectiveness its superiority comparative methods.

Language: Английский

---