Recent advances of Transformers in medical image analysis: A comprehensive review

MedComm – Future Medicine, Journal Year: 2023, Volume and Issue: 2(1)

Published: March 1, 2023

Abstract Recent works have shown that Transformer's excellent performances on natural language processing tasks can be maintained image analysis tasks. However, the complicated clinical settings in medical and varied disease properties bring new challenges for use of Transformer. The computer vision engineering communities devoted significant effort to research based Transformer with especial focus scenario‐specific architectural variations. In this paper, we comprehensively review rapidly developing area by covering latest advances Transformer‐based methods different settings. We first give introduction basic mechanisms including implementations selfattention typical architectures. important problems various data modalities, visual tasks, organs diseases are then reviewed systemically. carefully collect 276 very recent 76 public datasets an organized structure. Finally, discussions open future directions also provided. expect up‐to‐date roadmap serve as a reference source pursuit boosting development field.

Language: Английский

---

Semi-Supervised Learning of MRI Synthesis Without Fully-Sampled Ground Truths

IEEE Transactions on Medical Imaging, Journal Year: 2022, Volume and Issue: 41(12), P. 3895 - 3906

Published: Aug. 15, 2022

Learning-based translation between MRI contrasts involves supervised deep models trained using high-quality source- and target-contrast images derived from fully-sampled acquisitions, which might be difficult to collect under limitations on scan costs or time. To facilitate curation of training sets, here we introduce the first semi-supervised model for contrast (ssGAN) that can directly undersampled k-space data. enable learning data, ssGAN introduces novel multi-coil losses in image, k-space, adversarial domains. The are selectively enforced acquired samples unlike traditional single-coil synthesis models. Comprehensive experiments retrospectively multi-contrast brain datasets provided. Our results demonstrate yields par performance a model, while outperforming coil-combined magnitude images. It also outperforms cascaded reconstruction-synthesis where is following self-supervised reconstruction Thus, holds great promise improve feasibility learning-based synthesis.

Language: Английский

---

Anisotropic edge-preserving network for resolution enhancement in unidirectional Cartesian magnetic particle imaging

Physics in Medicine and Biology, Journal Year: 2023, Volume and Issue: 68(4), P. 045014 - 045014

Published: Jan. 23, 2023

Objective. Magnetic particle imaging (MPI) is a novel modality. It crucial to acquire accurate localization of the superparamagnetic iron oxide nanoparticles distributions in MPI. However, spatial resolution unidirectional Cartesian trajectory MPI exhibits anisotropy, which blurs boundaries images and makes precise difficult. In this paper, we propose an anisotropic edge-preserving network (AEP-net) alleviate MPI.Methods. AEP-net resolve anisotropy by constructing asymmertic convolution. To recover edge information, design uncertainty region module. addition, evaluated performance proposed model using simulations experimental data.Results. The results show that alleviates preserves details image. By comparing visualization metrics, demonstrate our method superior other methods.Significance. produces devices promotes quantization, promote biomedical applications

Language: Английский

---

COVID-19 Detection From Respiratory Sounds With Hierarchical Spectrogram Transformers

IEEE Journal of Biomedical and Health Informatics, Journal Year: 2023, Volume and Issue: 28(3), P. 1273 - 1284

Published: Dec. 5, 2023

Monitoring of prevalent airborne diseases such as COVID-19 characteristically involves respiratory assessments. While auscultation is a mainstream method for preliminary screening disease symptoms, its utility hampered by the need dedicated hospital visits. Remote monitoring based on recordings sounds portable devices promising alternative, which can assist in early assessment that primarily affects lower tract. In this study, we introduce novel deep learning approach to distinguish patients with from healthy controls given audio cough or breathing sounds. The proposed leverages hierarchical spectrogram transformer (HST) representations HST embodies self-attention mechanisms over local windows spectrograms, and window size progressively grown model stages capture global context. compared against state-of-the-art conventional deep-learning baselines. Demonstrations crowd-sourced multi-national datasets indicate outperforms competing methods, achieving 90% area under receiver operating characteristic curve (AUC) detecting cases.

Language: Английский

---

SCANeXt: Enhancing 3D medical image segmentation with dual attention network and depth-wise convolution

Heliyon, Journal Year: 2024, Volume and Issue: 10(5), P. e26775 - e26775

Published: Feb. 28, 2024

Existing approaches to 3D medical image segmentation can be generally categorized into convolution-based or transformer-based methods. While convolutional neural networks (CNNs) demonstrate proficiency in extracting local features, they encounter challenges capturing global representations. In contrast, the consecutive self-attention modules present vision transformers excel at long-range dependencies and achieving an expanded receptive field. this paper, we propose a novel approach, termed SCANeXt, for segmentation. Our method combines strengths of dual attention (Spatial Channel Attention) ConvNeXt enhance representation learning images. particular, mechanism crafted encompass spatial channel relationships throughout entire feature dimension. To further extract multiscale introduce depth-wise convolution block inspired by after block. Extensive evaluations on three benchmark datasets, namely Synapse, BraTS, ACDC, effectiveness our proposed terms accuracy. SCANeXt model achieves state-of-the-art result with Dice Similarity Score 95.18% ACDC dataset, significantly outperforming current

Language: Английский

---

Accurate Concentration Recovery for Quantitative Magnetic Particle Imaging Reconstruction via Nonconvex Regularization

IEEE Transactions on Medical Imaging, Journal Year: 2024, Volume and Issue: 43(8), P. 2949 - 2959

Published: April 1, 2024

Magnetic particle imaging (MPI) uses nonlinear response signals to noninvasively detect magnetic nanoparticles in space, and its quantitative properties hold promise for future precise treatments. In reconstruction, the system matrix based method necessitates suitable regularization terms, such as Tikhonov or non-negative fused lasso (NFL) regularization, stabilize solution. While NFL offers clearer edge information than it carries a biased estimate of l ₁ penalty, leading an underestimation reconstructed concentration adversely affecting properties. this paper, new nonconvex including min-max concave (MC) total variation (TV) is proposed. This utilized MC penalty provide nearly unbiased sparse constraints adds TV uniform intensity distribution images. By combining alternating direction multiplication (ADMM) two-step parameter selection method, more accurate MPI reconstruction was realized. The performance proposed verified on simulation data, Open-MPI dataset, measured data from homemade scanner. results indicate that achieves better image quality while maintaining properties, thus overcoming drawback by providing information. particular, reduced relative error 28% 8%.

Language: Английский

---

Progressively volumetrized deep generative models for data-efficient contextual learning of MR image recovery

Medical Image Analysis, Journal Year: 2022, Volume and Issue: 78, P. 102429 - 102429

Published: March 26, 2022

Language: Английский

---

Dual-Feature Frequency Component Compression Method for Accelerating Reconstruction in Magnetic Particle Imaging

IEEE Transactions on Computational Imaging, Journal Year: 2023, Volume and Issue: 9, P. 289 - 297

Published: Jan. 1, 2023

The frequency component compression method (FCCM) has been widely used in magnetic particle imaging (MPI) technology to improve reconstruction efficiency. This can reduce the time by using signal-to-noise ratio (SNR) feature remove high noise components. To further accelerate reconstruction, a dual-feature (DF-FCCM) was developed herein. A new energy spectral density (ESD) introduced describe level of measurement signal. By SNR and ESD feature, DF-FCCM select valuable components that contain low both signal system matrix. be reduced fewer more information. efficiency robustness proposed validated through extensive simulation experiments. Further real experiments based on OpenMPI data set verified applied MPI reconstruction. Compared previous SNR-FCCM, achieve similar or better quality 25% time. efficiently potential online imaging, which is essential for pre-clinical clinical applications.

Language: Английский

---

GCRDN: Global Context-Driven Residual Dense Network for Remote Sensing Image Superresolution

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Journal Year: 2023, Volume and Issue: 16, P. 4457 - 4468

Published: Jan. 1, 2023

Superresolution (SR) of remote sensing images aims to restore high-quality information from low-resolution images. Recently, it has witnessed great strides with the rapid development deep learning (DL) techniques. Despite their good performance, these DL-based models are often ineffective in balancing global and local feature extraction. Moreover, they usually hindered by poor image reconstruction capability decoder inside SR models. To cope this problem, work proposes a novel context-driven residual dense network (GCRDN) for satellite based on encoder architecture. In particular, proposed is endowed nonlocal sparse attention modules incorporated into learn robust representations features. Furthermore, equipped back-sampling blocks devised fully exploit maps extracted encoder. Extensive experimental comparisons two multisensor datasets confirm that GCRDN achieves impressive performance terms perceptual quality fidelity.

Language: Английский

---

A transformer-based hierarchical registration framework for multimodality deformable image registration

Computerized Medical Imaging and Graphics, Journal Year: 2023, Volume and Issue: 108, P. 102286 - 102286

Published: Aug. 10, 2023

Language: Английский

---