Simulated Quantum Mechanics-Based Joint Learning Network for Stroke Lesion Segmentation and TICI Grading DOI
Liangliang Liu, Chang Jing, Gongbo Liang

et al.

IEEE Journal of Biomedical and Health Informatics, Journal Year: 2023, Volume and Issue: 27(7), P. 3372 - 3383

Published: April 27, 2023

Segmenting stroke lesions and assessing the thrombolysis in cerebral infarction (TICI) grade are two important but challenging prerequisites for an auxiliary diagnosis of stroke. However, most previous studies have focused only on a single one tasks, without considering relation between them. In our study, we propose simulated quantum mechanics-based joint learning network (SQMLP-net) that simultaneously segments lesion assesses TICI grade. The correlation heterogeneity tasks tackled with single-input double-output hybrid network. SQMLP-net has segmentation branch classification branch. These branches share encoder, which extracts shares spatial global semantic information tasks. Both optimized by novel loss function learns intra- inter-task weights these Finally, evaluate public dataset (ATLAS R2.0). obtains state-of-the-art metrics (Dice:70.98% accuracy:86.78%) outperforms single-task existing advanced methods. An analysis found negative severity grading accuracy segmentation.

Language: Английский

A survey on deep learning tools dealing with data scarcity: definitions, challenges, solutions, tips, and applications DOI Creative Commons
Laith Alzubaidi, Jinshuai Bai, Aiman Al-Sabaawi

et al.

Journal Of Big Data, Journal Year: 2023, Volume and Issue: 10(1)

Published: April 14, 2023

Abstract Data scarcity is a major challenge when training deep learning (DL) models. DL demands large amount of data to achieve exceptional performance. Unfortunately, many applications have small or inadequate train frameworks. Usually, manual labeling needed provide labeled data, which typically involves human annotators with vast background knowledge. This annotation process costly, time-consuming, and error-prone. every framework fed by significant automatically learn representations. Ultimately, larger would generate better model its performance also application dependent. issue the main barrier for dismissing use DL. Having sufficient first step toward any successful trustworthy application. paper presents holistic survey on state-of-the-art techniques deal models overcome three challenges including small, imbalanced datasets, lack generalization. starts listing techniques. Next, types architectures are introduced. After that, solutions address listed, such as Transfer Learning (TL), Self-Supervised (SSL), Generative Adversarial Networks (GANs), Model Architecture (MA), Physics-Informed Neural Network (PINN), Deep Synthetic Minority Oversampling Technique (DeepSMOTE). Then, these were followed some related tips about acquisition prior purposes, well recommendations ensuring trustworthiness dataset. The ends list that suffer from scarcity, several alternatives proposed in order more each Electromagnetic Imaging (EMI), Civil Structural Health Monitoring, Medical imaging, Meteorology, Wireless Communications, Fluid Mechanics, Microelectromechanical system, Cybersecurity. To best authors’ knowledge, this review offers comprehensive overview strategies tackle

Language: Английский

Citations

366
Vision Transformers in medical computer vision—A contemplative retrospection DOI

Arshi Parvaiz,

Muhammad Anwaar Khalid,

Rukhsana Zafar

et al.

Engineering Applications of Artificial Intelligence, Journal Year: 2023, Volume and Issue: 122, P. 106126 - 106126

Published: March 20, 2023

Language: Английский

Citations

152
AGGN: Attention-based glioma grading network with multi-scale feature extraction and multi-modal information fusion DOI
Peishu Wu, Zidong Wang,

Baixun Zheng

et al.

Computers in Biology and Medicine, Journal Year: 2022, Volume and Issue: 152, P. 106457 - 106457

Published: Dec. 21, 2022

Language: Английский

Citations

109
Deep multimodal fusion of image and non-image data in disease diagnosis and prognosis: a review DOI Creative Commons
Can Cui,

Haichun Yang,

Yaohong Wang

et al.

Progress in Biomedical Engineering, Journal Year: 2023, Volume and Issue: 5(2), P. 022001 - 022001

Published: March 9, 2023

Abstract The rapid development of diagnostic technologies in healthcare is leading to higher requirements for physicians handle and integrate the heterogeneous, yet complementary data that are produced during routine practice. For instance, personalized diagnosis treatment planning a single cancer patient relies on various images (e.g. radiology, pathology camera images) non-image clinical genomic data). However, such decision-making procedures can be subjective, qualitative, have large inter-subject variabilities. With recent advances multimodal deep learning technologies, an increasingly number efforts been devoted key question: how do we extract aggregate information ultimately provide more objective, quantitative computer-aided decision making? This paper reviews studies dealing with question. Briefly, this review will include (a) overview current workflows, (b) summarization fusion methods, (c) discussion performance, (d) applications disease prognosis, (e) challenges future directions.

Language: Английский

Citations

85
Multi-task deep learning for medical image computing and analysis: A review DOI
Yan Zhao, Xiuying Wang, Tongtong Che

et al.

Computers in Biology and Medicine, Journal Year: 2022, Volume and Issue: 153, P. 106496 - 106496

Published: Dec. 28, 2022

Language: Английский

Citations

83
Emerging role of artificial intelligence in diagnosis, classification and clinical management of glioma DOI
Jiefeng Luo, Mika Pan,

Ke Mo

et al.

Seminars in Cancer Biology, Journal Year: 2023, Volume and Issue: 91, P. 110 - 123

Published: March 11, 2023

Language: Английский

Citations

53
A review of deep learning-based information fusion techniques for multimodal medical image classification DOI Creative Commons
Yihao Li, Mostafa El Habib Daho, Pierre-Henri Conze

et al.

Computers in Biology and Medicine, Journal Year: 2024, Volume and Issue: 177, P. 108635 - 108635

Published: May 22, 2024

Multimodal medical imaging plays a pivotal role in clinical diagnosis and research, as it combines information from various modalities to provide more comprehensive understanding of the underlying pathology. Recently, deep learning-based multimodal fusion techniques have emerged powerful tools for improving image classification. This review offers thorough analysis developments classification tasks. We explore complementary relationships among prevalent outline three main schemes networks: input fusion, intermediate (encompassing single-level hierarchical attention-based fusion), output fusion. By evaluating performance these techniques, we insight into suitability different network architectures scenarios application domains. Furthermore, delve challenges related architecture selection, handling incomplete data management, potential limitations Finally, spotlight promising future Transformer-based give recommendations research this rapidly evolving field.

Language: Английский

Citations

24
Hybrid CNN-Transformer Network With Circular Feature Interaction for Acute Ischemic Stroke Lesion Segmentation on Non-Contrast CT Scans DOI
Hulin Kuang, Yahui Wang, Jin Liu

et al.

IEEE Transactions on Medical Imaging, Journal Year: 2024, Volume and Issue: 43(6), P. 2303 - 2316

Published: Feb. 6, 2024

Lesion segmentation is a fundamental step for the diagnosis of acute ischemic stroke (AIS). Non-contrast CT (NCCT) still mainstream imaging modality AIS lesion measurement. However, on NCCT challenging due to low contrast, noise and artifacts. To achieve accurate NCCT, this study proposes hybrid convolutional neural network (CNN) Transformer with circular feature interaction bilateral difference learning. It consists parallel CNN encoders, module, shared decoder learning module. A new block particularly designed solve weak inductive bias problem traditional Transformer. effectively combine features from we first design multi-level aggregation module multi-scale in each encoder then propose novel containing CNN-to-Transformer Transformer-to-CNN blocks. Besides, proposed at bottom level learn different information between contralateral sides brain. The method evaluated three datasets: public AISD, private dataset an external dataset. Experimental results show that achieves Dices 61.39% 46.74% AISD dataset, respectively, outperforming 17 state-of-the-art methods. volumetric analysis segmented lesions validation imply potential provide support diagnosis.

Language: Английский

Citations

19
Advancements in medical image segmentation: A review of transformer models DOI

S. S. Kumar

Computers & Electrical Engineering, Journal Year: 2025, Volume and Issue: 123, P. 110099 - 110099

Published: Jan. 22, 2025

Language: Английский

Citations

2
Global Transformer and Dual Local Attention Network via Deep-Shallow Hierarchical Feature Fusion for Retinal Vessel Segmentation DOI
Yang Li, Yue Zhang, Jingyu Liu

et al.

IEEE Transactions on Cybernetics, Journal Year: 2022, Volume and Issue: 53(9), P. 5826 - 5839

Published: Aug. 19, 2022

Clinically, retinal vessel segmentation is a significant step in the diagnosis of fundus diseases. However, recent methods generally neglect difference semantic information between deep and shallow features, which fail to capture global local characterizations images simultaneously, resulting limited performance for fine vessels. In this article, transformer (GT) dual attention (DLA) network via deep-shallow hierarchical feature fusion (GT-DLA-dsHFF) are investigated solve above limitations. First, GT developed integrate image, effectively captures long-distance dependence pixels, alleviating discontinuity blood vessels results. Second, DLA, constructed using dilated convolutions with varied dilation rates, unsupervised edge detection, squeeze-excitation block, proposed extract information, consolidating details result. Finally, novel (dsHFF) algorithm studied fuse features different scales learning framework, respectively, can mitigate attenuation valid process fusion. We verified GT-DLA-dsHFF on four typical image datasets. The experimental results demonstrate our achieves superior against current detailed discussions verify efficacy three modules. Segmentation diseased show robustness GT-DLA-dsHFF. Implementation codes will be available https://github.com/YangLibuaa/GT-DLA-dsHFF.

Language: Английский

Citations

66