Related Experiment Video
Updated: Apr 1, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.7K
Deep Learning for Medical Ultrasound Image Segmentation: A Systematic Review of the Current Research.
Oona Rainio1, Ehsan Roshan2,3, Seyed Mohammedreza Hosseini2
1Turku PET Centre, University of Turku and Turku University Hospital, Turku, Finland. ormrai@utu.fi.
Journal of Imaging Informatics in Medicine
|March 30, 2026
Summary
Deep learning models excel at segmenting ultrasound images, particularly for breast tumors and organs. While newer models like vision transformers emerge, convolutional neural networks remain popular, with no clear performance advantage linked to model type.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Deep learning (DL) has revolutionized medical image analysis, enabling automated segmentation.
- Ultrasound imaging is increasingly utilized across various medical specialties.
- A comprehensive overview of DL applications in ultrasound segmentation is needed due to rapid advancements.
Purpose of the Study:
- To systematically review and analyze the current research landscape of deep learning-based ultrasound image segmentation.
- To identify common segmentation targets, popular DL architectures, and emerging trends.
Main Methods:
- Systematic literature review following PRISMA 2020 guidelines.
- Analysis of 296 scientific articles from the PubMed database.
- Categorization of segmentation targets and DL model architectures.
Main Results:
- Common segmentation targets include breast tumors, organs, and cardiovascular structures, with applications in orthopedics, thyroid nodules, and oncology.
- Convolutional Neural Networks (CNNs), particularly U-shaped architectures, remain popular, alongside emerging Vision Transformers (ViTs) and Segment Anything Models.
- No significant association was found between DL model type and reported evaluation metrics, despite newer models using more data.
- Key limitations include lack of computational requirement data and inconsistent performance evaluation.
Conclusions:
- Deep learning-based ultrasound segmentation is a rapidly advancing field driven by increased ultrasound use, new datasets, and methodological progress.
- Further standardization in reporting computational costs and performance evaluation is crucial for future research.
- The field shows promise for improved diagnostic accuracy and clinical workflows in diverse medical areas.
