Related Experiment Video
Updated: Jul 21, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.9K
A Deep Learning Framework for the Characterization of Thyroid Nodules from Ultrasound Images Using Improved Inception
O A Ajilisa1, V P Jagathy Raj2, M K Sabu1
1Department of Computer Applications, Cochin University of Science and Technology, South Kalamassery, Kochi 682022, Kerala, India.
Diagnostics (Basel, Switzerland)
|July 29, 2023
Summary
This study introduces a deep learning framework using improved inception blocks and multi-level transfer learning for thyroid nodule diagnosis from ultrasound images. The model achieves high accuracy, aiding radiologists and potentially reducing unnecessary procedures.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Deep learning models require substantial data for accurate diagnosis of thyroid nodules.
- Acquiring sufficient medical images for training deep learning models remains a significant challenge.
Purpose of the Study:
- To develop a deep learning framework for differentiating benign and malignant thyroid nodules using ultrasound images.
- To enhance diagnostic accuracy and assist radiologists in clinical decision-making, potentially reducing unnecessary fine-needle aspirations.
Main Methods:
- Developed a deep learning framework integrating squeeze and excitation networks with inception modules for improved recognition accuracy.
- Implemented multi-level transfer learning using breast ultrasound images as a bridge dataset to address domain differences.
- Utilized improved inception blocks and multi-level transfer learning for thyroid nodule characterization.
Main Results:
- Achieved high precision (0.9057 benign, 0.9667 malignant), recall (0.9796 benign, 0.8529 malignant), and F1-score (0.9412 benign, 0.9062 malignant).
- Obtained an AUC value of 0.9537, outperforming single-level transfer learning.
- Demonstrated classification accuracy comparable to experienced radiologists.
Conclusions:
- The proposed deep learning framework effectively differentiates thyroid nodules with high accuracy.
- The model shows potential for clinical application, offering time and effort savings for radiologists.
- This approach can aid in avoiding unnecessary fine-needle aspirations for benign thyroid nodules.
