Related Experiment Video
Updated: Aug 3, 2025
05:56
Objectification of Tongue Diagnosis in Traditional Medicine, Data Analysis, and Study Application
Published on: April 14, 2023
2.6K
Multi-path decoder U-Net: A weakly trained real-time segmentation network for object detection and localization in
Abdullah F Al-Battal1, Imanuel R Lerman2, Truong Q Nguyen3
1Electrical and Computer Engineering Department, University of California, San Diego, CA 92093, USA; Electrical Engineering Department, King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia.
Summary
This study introduces a new U-Net deep learning model for ultrasound image analysis. It efficiently detects anatomical structures using bounding boxes, reducing annotation costs and improving diagnostic accuracy.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Ultrasound Diagnostics
Background:
- Accurate detection and localization of anatomical structures in ultrasound scans are crucial for medical procedures.
- Variability in ultrasound images due to sonographers and patients complicates accurate identification.
- Current deep learning models (CNNs) require extensive pixel-wise annotations, increasing costs and time.
Purpose of the Study:
- To develop a more efficient deep learning model for anatomical structure detection in ultrasound images.
- To reduce the reliance on costly and time-consuming pixel-wise annotations for training.
- To improve the accuracy and efficiency of object detection and localization in ultrasound scans.
Main Methods:
- Proposed a novel multi-path decoder U-Net architecture.
- Trained the network using bounding box segmentation maps instead of pixel-wise annotations.
- Evaluated the model's performance on small medical imaging datasets.
Main Results:
- The proposed U-Net architecture achieved up to a 7% relative improvement in localization and detection performance compared to standard U-Net.
- The model requires minimal training data, making it suitable for medical imaging applications.
- Performance was comparable or superior to U-Net++, a more computationally expensive model.
Conclusions:
- The multi-path decoder U-Net offers a computationally efficient and accurate solution for real-time object detection and localization in ultrasound scans.
- This approach significantly reduces the cost and time associated with training deep learning models for medical imaging.
- The model's ability to train on limited data makes it a practical tool for clinical settings.
