4.6 Article Proceedings Paper

Stacked fully convolutional networks with multi-channel learning: application to medical image segmentation

期刊

VISUAL COMPUTER
卷 33, 期 6-8, 页码 1061-1071

出版社

SPRINGER
DOI: 10.1007/s00371-017-1379-4

关键词

Fully convolutional networks (FCNs); Segmentation; Regions of interest (ROI)

向作者/读者索取更多资源

The automated segmentation of regions of interest (ROIs) in medical imaging is the fundamental requirement for the derivation of high-level semantics for image analysis in clinical decision support systems. Traditional segmentation approaches such as region-based depend heavily upon hand-crafted features and a priori knowledge of the user. As such, these methods are difficult to adopt within a clinical environment. Recently, methods based on fully convolutional networks (FCN) have achieved great success in the segmentation of general images. FCNs leverage a large labeled dataset to hierarchically learn the features that best correspond to the shallow appearance as well as the deep semantics of the images. However, when applied to medical images, FCNs usually produce coarse ROI detection and poor boundary definitions primarily due to the limited number of labeled training data and limited constraints of label agreement among neighboring similar pixels. In this paper, we propose a new stacked FCN architecture with multi-channel learning (SFCN-ML). We embed the FCN in a stacked architecture to learn the foreground ROI features and background non-ROI features separately and then integrate these different channels to produce the final segmentation result. In contrast to traditional FCN methods, our SFCN-ML architecture enables the visual attributes and semantics derived from both the fore- and background channels to be iteratively learned and inferred. We conducted extensive experiments on three public datasets with a variety of visual challenges. Our results show that our SFCN-ML is more effective and robust than a routine FCN and its variants, and other state-of-the-art methods.

作者

我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。

评论

主要评分

4.6
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

Article Clinical Neurology

Biomarker clustering in autosomal dominant Alzheimer's disease

Patrick H. Luckett, Charlie Chen, Brian A. Gordon, Julie Wisch, Sarah B. Berman, Jasmeer P. Chhatwal, Carlos Cruchaga, Anne M. Fagan, Martin R. Farlow, Nick C. Fox, Mathias Jucker, Johannes Levin, Colin L. Masters, Hiroshi Mori, James M. Noble, Stephen Salloway, Peter R. Schofield, Adam M. Brickman, William S. Brooks, David M. Cash, Michael J. Fulham, Bernardino Ghetti, Clifford R. Jack, Jonathan Voeglein, William E. Klunk, Robert Koeppe, Yi Su, Michael Weiner, Qing Wang, Daniel Marcus, Deborah Koudelis, Nelly Joseph-Mathurin, Lisa Cash, Russ Hornbeck, Chengjie Xiong, Richard J. Perrin, Celeste M. Karch, Jason Hassenstab, Eric McDade, John C. Morris, Tammie L. S. Benzinger, Randall J. Bateman, Beau M. Ances

Summary: This study analyzed 19 biomarkers of Alzheimer's disease using hierarchical clustering and feature selection, and found that amyloid and tau measures were the primary predictors. Emerging biomarkers of neuronal integrity and inflammation showed weaker predictive ability.

ALZHEIMERS & DEMENTIA (2023)

Editorial Material Clinical Neurology

Erdheim-Chester disease presenting as precipitous cognitive decline

Sophie Dunkerton, Ross Penninkilampi, Heidi Beadnall, Michael Fulham, Andrew Colebatch, Stacey Jankelowitz, Rebekah Ahmed, Zoe Thayer, Michael Halmagyi, Edward Abadir

PRACTICAL NEUROLOGY (2023)

Article Computer Science, Interdisciplinary Applications

Graph-Based Intercategory and Intermodality Network for Multilabel Classification and Melanoma Diagnosis of Skin Lesions in Dermoscopy and Clinical Images

Xiaohang Fu, Lei Bi, Ashnil Kumar, Michael Fulham, Jinman Kim

Summary: The identification of melanoma can be done through the analysis of clinical and dermoscopy images. Current methods lack the ability to fully utilize information from both modalities and exploit the intercategory relationships in the 7PC. This study proposes a graph-based network with two modules to address these limitations and improves classification performance.

IEEE TRANSACTIONS ON MEDICAL IMAGING (2022)

Article Biology

Deep multimodal graph-based network for survival prediction from highly multiplexed images and patient variables

Xiaohang Fu, Ellis Patrick, Jean Y. H. Yang, David Dagan Feng, Jinman Kim

Summary: The spatial architecture and phenotypic heterogeneity of tumor cells are associated with cancer prognosis and outcomes. Imaging mass cytometry captures high-dimensional maps of disease-relevant biomarkers at single-cell resolution, which can inform patient-specific prognosis. However, existing methods for survival prediction do not utilize spatial phenotype information at the single-cell level, and there is a lack of end-to-end methods that integrate imaging data with clinical information for improved accuracy. We propose a deep multimodal graph-based network that considers spatial phenotype information and clinical variables to enhance survival prediction, and demonstrate its effectiveness in breast cancer datasets.

COMPUTERS IN BIOLOGY AND MEDICINE (2023)

Article Computer Science, Information Systems

Vision-Language Transformer for Interpretable Pathology Visual Question Answering

Usman Naseem, Matloob Khushi, Jinman Kim

Summary: Pathology visual question answering (PathVQA) aims to answer medical questions using pathology images. Existing methods have limitations in capturing the high and low-level interactions between vision and language features required for VQA. Additionally, these methods lack interpretability in justifying the retrieved answers. To address these limitations, a vision-language transformer called TraP-VQA is introduced, which embeds vision and language features for interpretable PathVQA. Our experiments demonstrate that TraP-VQA outperforms state-of-the-art methods and validate its robustness on medical VQA datasets, along with the capability of the integrated vision-language model. Visualization results explain the reasoning behind the retrieved PathVQA answers.

IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS (2023)

Article Automation & Control Systems

Unsupervised Landmark Detection-Based Spatiotemporal Motion Estimation for 4-D Dynamic Medical Images

Yuyu Guo, Lei Bi, Dongming Wei, Liyun Chen, Zhengbin Zhu, Dagan Feng, Ruiyan Zhang, Qian Wang, Jinman Kim

Summary: In this study, we propose a dense-sparse-dense (DSD) motion estimation framework that utilizes unsupervised 3D landmark detection network and motion reconstruction network to extract sparse landmarks and construct motion field in two stages. The method improves the accuracy of motion estimation and preserves anatomical topology.

IEEE TRANSACTIONS ON CYBERNETICS (2023)

Letter Hematology

Lenalidomide Consolidation Added to Rituximab Maintenance Therapy in Patients Remaining PET Positive After Treatment for Relapsed Follicular Lymphoma: A Phase 2 Australasian Leukaemia & Lymphoma Group NHL26 Study

Judith Trotman, Peter Presgrave, Duncan P. Carradice, Douglas Stuart Lenton, Maher K. Gandhi, Tara Cochrane, Xavier Badoux, Julia Carlson, Gloria Nkhoma, Belinda Butcher, Armin Nikpour, Michael Fulham, Anna M. Johnston

HEMASPHERE (2023)

Article Computer Science, Interdisciplinary Applications

A Shortened Model for Logan Reference Plot Implemented via the Self-Supervised Neural Network for Parametric PET Imaging

Wenxiang Ding, Qiaoqiao Ding, Kewei Chen, Miao Zhang, Li Lv, David Dagan Feng, Lei Bi, Jinman Kim, Qiu Huang

Summary: Dynamic PET imaging provides more comprehensive physiological information than conventional static PET imaging. The proposed modified Logan reference plot model and self-supervised convolutional neural network improve noise performance and accurately estimate the distribution volume ratio in dynamic PET with a shortened scanning protocol. The method has the potential to add clinical value by providing both DVR and SUV simultaneously.

IEEE TRANSACTIONS ON MEDICAL IMAGING (2023)

Article Computer Science, Cybernetics

RHMD: A Real-World Dataset for Health Mention Classification on Reddit

Usman Naseem, Matloob Khushi, Jinman Kim, Adam G. Dunn

Summary: People on social media using disease and symptom words to discuss their health can introduce biases in data-driven public health applications. This study presents a new dataset called RHMD, which consists of 10,015 manually annotated Reddit posts. The dataset is labeled with four categories and provides a comprehensive performance analysis of baseline methods. The release of this dataset is expected to facilitate the development of new methods for detecting health mentions in user-generated text.

IEEE TRANSACTIONS ON COMPUTATIONAL SOCIAL SYSTEMS (2023)

Review Automation & Control Systems

A Review of Predictive and Contrastive Self-supervised Learning for Medical Images

Wei-Chien Wang, Euijoon Ahn, Dagan Feng, Jinman Kim

Summary: Over the last decade, supervised deep learning has made significant progress in computer vision tasks using manually annotated big data. However, the limited availability of high-quality annotated medical imaging data hinders the application of deep learning in medical image analysis. A potential solution is the use of self-supervised learning (SSL), particularly contrastive SSL, which has shown promise in rivaling or surpassing supervised learning. This review examines state-of-the-art contrastive SSL algorithms originally designed for natural images, explores their adaptations for medical images, and discusses recent advances, current limitations, and future directions in applying contrastive SSL in the medical domain.

MACHINE INTELLIGENCE RESEARCH (2023)

Article Biochemistry & Molecular Biology

Synthesis and structure-activity relationship (SAR) studies of 1,2,3-triazole, amide, and ester-based benzothiazole derivatives as potential molecular probes for tau protein

Hendris Wongso, Maiko Ono, Tomoteru Yamasaki, Katsushi Kumata, Makoto Higuchi, Ming-Rong Zhang, Michael J. Fulham, Andrew Katsifis, Paul A. Keller

Summary: The pyridinyl-butadienyl-benzothiazole (PBB3 15) scaffold was used to improve tau ligands for imaging Alzheimer's disease. Triazole derivatives visualized A beta plaques but failed to detect neurofibrillary tangles (NFTs), while amide 110 and ester 129 successfully observed NFTs. These ligands showed different affinities at the binding sites with PBB3.

RSC MEDICINAL CHEMISTRY (2023)

Article Computer Science, Cybernetics

Hybrid Text Representation for Explainable Suicide Risk Identification on Social Media

Usman Naseem, Matloob Khushi, Jinman Kim, Adam G. Dunn

Summary: The article introduces a hybrid text representation method for explaining suicide risk identification on social media. The method achieves excellent results on a public suicide dataset and demonstrates advantages in clinical and public health practice.

IEEE TRANSACTIONS ON COMPUTATIONAL SOCIAL SYSTEMS (2022)

Article Computer Science, Artificial Intelligence

SparseVoxNet: 3-D Object Recognition With Sparsely Aggregation of 3-D Dense Blocks

Ahmad Karambakhsh, Bin Sheng, Ping Li, Huating Li, Jinman Kim, Younhyun Jung, C. L. Philip Chen

Summary: The article introduces a novel solution for 3-D object recognition from volumetric data by combining three compact CNN models, low-cost SparseNet, and feature representation technique. By estimating extra geometrical information, an optimized network is achieved and improves the recognition results.

IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS (2022)

暂无数据