Related Experiment Video
Updated: Jun 9, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.8K
Automatic tumor segmentation and lymph node metastasis prediction in papillary thyroid carcinoma using ultrasound
Xian-Ya Zhang1, Di Zhang2, Zhi-Yuan Wang3
1Department of Medical Ultrasound, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China.
Medical Physics
|October 30, 2024
Summary
A new deep learning model accurately predicts lymph node metastasis in papillary thyroid cancer using ultrasound video keyframes. This automated tool improves upon existing methods and assists radiologists, enhancing patient management.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Oncology and Radiology
Background:
- Accurate preoperative prediction of cervical lymph node metastasis (LNM) is crucial for papillary thyroid carcinoma (PTC) staging and treatment.
- Improved prognosis and management depend on precise LNM assessment.
Purpose of the Study:
- To develop a fully automated deep learning-enabled model (FADLM) for tumor segmentation and LNM prediction in PTC.
- Utilize ultrasound (US) video keyframes for automated analysis.
Main Methods:
- Developed FADLM integrating Mask R-CNN for segmentation and ResNet34 for LNM diagnosis.
- Compared FADLM against radiomics (RM, TRM) and clinical-semantic models (CSM).
- Validated performance using AUC, heatmap, and DCA; assessed radiologist performance with FADLM assistance.
Main Results:
- FADLM achieved high segmentation performance (DSC: 0.85–0.88) across cohorts.
- FADLM demonstrated superior LNM prediction AUCs (0.78–0.83) compared to RM and CSM.
- FADLM significantly improved radiologist accuracy and sensitivity, reducing segmentation time.
Conclusions:
- The FADLM shows strong potential for efficient and consistent cervical LNM prediction in PTC using US video keyframes.
- FADLM outperforms existing models and human radiologists, offering promising clinical efficacy.
Keywords:
cervical lymph node metastasisdeep learningfully automated modelthyroid papillary carcinomaultrasound
