A Tumor Segmentation Method Based on Mean-Teacher Reusing Pseudo-Labels DOI Creative Commons
Chengyu Jiang,

Shangkun Liu,

Jingyu Wang

et al.

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 41942 - 41953

Published: Jan. 1, 2024

Breast tumor is a common female physiological disease, and the malignant one of main fatal diseases women. Accurate examination assessment shape can facilitate subsequent treatment improve cure rate. With development deep learning, automatic detection systems are designed to assist doctors in diagnosis. However, blurry edges, poor visual quality, irregular shapes breast tumors pose significant challenges design highly efficient system. In addition, lack publicly available labeled data major obstacle developing accurate robust learning models for detection. To overcome aforementioned issues, we propose SRU-PMT+, pseudo-label reusing Mean-Teacher architecture based on squeeze-and-excitation residual (SE-Res) attention. We utilize proposed segmentation network, SRU-Net++, generate pseudo-labels unlabeled data, guide student model using generated groundtruth, improving accuracy robustness model. Our semi-supervised method has been rigorously evaluated dataset, i.e., Ultrasound Images (BUSI) dataset. Results show that our outperforms current methods good performance. Importantly, strategy improves performance segmentation.

Language: Английский

Asym-UNet: An asymmetric U-shape Network for breast lesions ultrasound images segmentation DOI
Jia Liu, Jun Shao, Sen Xu

et al.

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 99, P. 106822 - 106822

Published: Sept. 12, 2024

Language: Английский

Citations

2
Performance Comparison ConvDeconvNet Algorithm Vs. UNET for Fish Object Detection DOI Creative Commons
Djarot Hindarto

SinkrOn, Journal Year: 2023, Volume and Issue: 8(4), P. 2827 - 2835

Published: Nov. 2, 2023

The precise identification and localization of fish entities within visual data is essential in diverse domains, such as marine biology fisheries management, computer vision. This study provides a thorough performance evaluation two prominent deep learning algorithms, ConvDeconvNet UNET, the context object detection. Both models are assessed using dataset comprising wide range species, considering various factors, including accuracy detection, speed processing, complexity model. findings demonstrate that exhibits superior terms detection accuracy, attaining noteworthy degree precision recall identifying entities. In contrast, UNET model displays notable advantage processing owing to its distinctive architectural design, rendering it viable option for applications requiring real-time performance. discourse surrounding trade-off between examined, offering valuable perspectives algorithm selection following specific criteria. Furthermore, this highlights significance incorporating datasets training testing purposes when utilizing these models, significantly influences their overall makes contribution continuous endeavors improve objects underwater images. It comparison thereby assisting researchers practitioners making well-informed decisions regarding selecting applications.

Language: Английский

Citations

4
Towards semi-supervised multi-modal rectal cancer segmentation: A large-scale dataset and a multi-teacher uncertainty-aware network DOI
Yu Qiu, Haotian Lu, Jie Mei

et al.

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: 255, P. 124734 - 124734

Published: July 10, 2024

Language: Английский

Citations

1
End-to-end deep learning pipeline for on-board extraterrestrial rock segmentation DOI Creative Commons

Daniel Marek,

Jakub Nalepa

Engineering Applications of Artificial Intelligence, Journal Year: 2023, Volume and Issue: 127, P. 107311 - 107311

Published: Oct. 24, 2023

Bringing autonomy on board edge devices is inevitable to accelerate the process of space exploration. Although there are various tasks that can be executed autonomously by such vehicles, detecting and segmenting rocks in on-board images extraterrestrial landscapes a critical step processing chain, as it allow navigate safely while avoiding collisions. We tackle this issue introduce an end-to-end pipeline for building validating resource-frugal machine learning techniques task, offering high level flexibility. Deploying models poses numerous practical challenges, spanning across ensuring their memory computational efficiency, understanding robustness against varying quality acquired images. These aspects often overlooked deep learning-powered systems—we show they (and ultimately should) part deployment chain. Our extensive experimental study performed over several benchmark data sets shed more light functional non-functional capabilities investigated models, both full-precision compressed quantisation, latter delivering statistically same segmentation accuracy being approximately 11× smaller. Additionally, we synthesised utilised quantify acquisition conditions which directly affect captured images—such simulations mimicking real-world settings could have negative impact trained clean high-quality image data. To ensure full reproducibility study, made our implementation publicly available at https://github.com/danielmarek22/onboard-rock-segmentation.

Language: Английский

Citations

3
A Tumor Segmentation Method Based on Mean-Teacher Reusing Pseudo-Labels DOI Creative Commons
Chengyu Jiang,

Shangkun Liu,

Jingyu Wang

et al.

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 41942 - 41953

Published: Jan. 1, 2024

Breast tumor is a common female physiological disease, and the malignant one of main fatal diseases women. Accurate examination assessment shape can facilitate subsequent treatment improve cure rate. With development deep learning, automatic detection systems are designed to assist doctors in diagnosis. However, blurry edges, poor visual quality, irregular shapes breast tumors pose significant challenges design highly efficient system. In addition, lack publicly available labeled data major obstacle developing accurate robust learning models for detection. To overcome aforementioned issues, we propose SRU-PMT+, pseudo-label reusing Mean-Teacher architecture based on squeeze-and-excitation residual (SE-Res) attention. We utilize proposed segmentation network, SRU-Net++, generate pseudo-labels unlabeled data, guide student model using generated groundtruth, improving accuracy robustness model. Our semi-supervised method has been rigorously evaluated dataset, i.e., Ultrasound Images (BUSI) dataset. Results show that our outperforms current methods good performance. Importantly, strategy improves performance segmentation.

Language: Английский

Citations

0