Related Experiment Video
Updated: Aug 6, 2026
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
A Generative Multi-modal Deep Learning Framework for Non-invasive Prediction of Ki-67 Expression in Papillary Thyroid
IEEE Journal of Biomedical and Health Informatics
|July 16, 2026
Summary
This study introduces a novel AI framework for non-invasively predicting Ki-67 expression in papillary thyroid microcarcinoma using B-mode ultrasound. The method enhances diagnostic accuracy without invasive biopsies, crucial for small nodules.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Accurate Ki-67 assessment is vital for papillary thyroid microcarcinoma (PTMC) prognosis and risk stratification.
- Current clinical evaluation relies on invasive biopsy, posing limitations for small nodules.
Purpose of the Study:
- To develop a non-invasive method for predicting Ki-67 expression using only preoperative B-mode ultrasound.
- To create a generative multi-modal and multi-task learning framework for enhanced diagnostic capabilities.
Main Methods:
- A BUS2USE network synthesized virtual ultrasound elastography from B-mode images.
- A reliability-aware fusion mechanism balanced real and synthetic features, enhanced by a multi-scale attention refinement module (MARM).
- VEGF prediction was incorporated as an auxiliary task for multi-task learning.
Main Results:
- The proposed framework achieved an accuracy of 0.852 and an AUC of 0.889.
- It outperformed existing Convolutional Neural Network (CNN) and transformer-based models.
- Analyses confirmed the framework's reliability and robustness.
Conclusions:
- The integration of synthetic imaging, attention-based fusion, and multi-task supervision enables effective non-invasive Ki-67 prediction.
- This AI-driven approach offers a valuable alternative to invasive testing, especially for small thyroid nodules.
