Related Experiment Video
Updated: Jul 12, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.9K
[An Improved Object Detection Algorithm for Thyroid Nodule Ultrasound Image Based on Faster R-CNN].
Tianlei Zheng1,2, Na Yang2, Shi Geng2
1School of Information and Control Engineering, China University of Mining and Technology, Xuzhou 221116, China.
Summary
This study introduces an enhanced Faster R-CNN algorithm for precise thyroid nodule detection in ultrasound images. The improved model significantly boosts accuracy and recall rates for identifying thyroid nodules.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Context:
- Thyroid nodules require accurate detection in ultrasound images for timely diagnosis.
- Existing object detection algorithms face challenges with nodule variability and size.
- Faster R-CNN is a foundational model for object detection tasks.
Purpose:
- To enhance the precision of thyroid nodule detection in ultrasound images.
- To improve the identification of irregularly shaped and small thyroid nodules.
- To increase the overall accuracy and recall rate of thyroid nodule detection algorithms.
Summary:
- An improved Faster R-CNN algorithm utilizes ResNeSt50 with deformable convolution for enhanced nodule feature extraction.
- Feature pyramid networks and RoI Align address missed detections and improve small nodule precision.
- Sharpness-Aware Minimization (SAM) enhances model generalization, achieving a 97.4% AP50 and a 10% improvement in AP@50:5:95.
Impact:
- The developed algorithm demonstrates superior accuracy and precision in thyroid nodule detection compared to original and existing models.
- Achieves a higher recall rate while maintaining lower detection frame precision requirements.
- Offers a robust and effective tool for clinical application in thyroid ultrasound analysis.
