Related Experiment Video
Updated: Sep 3, 2025
05:41
Author Spotlight: Integrating Ultrasound Imaging with Biochemical Markers for Thyroid Disease Diagnosis
Published on: February 9, 2024
723
Tracked 3D ultrasound and deep neural network-based thyroid segmentation reduce interobserver variability in thyroid
Markus Krönke1,2, Christine Eilers3, Desislava Dimova3
1Department of Radiology and Nuclear Medicine, German Heart Center, Technical University of Munich, Munich, Germany.
Plos One
|July 29, 2022
Summary
Tracked 3D ultrasound with deep neural network segmentation significantly improves thyroid volumetry accuracy and reduces operator variability compared to 2D ultrasound. This advanced method offers faster acquisition times and more reliable measurements for thyroid disease management.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Endocrinology
Background:
- Thyroid volumetry is essential for diagnosing and managing thyroid conditions.
- Traditional 2D ultrasound methods for thyroid volumetry are prone to significant operator dependency.
- Improving the accuracy and reproducibility of thyroid volume measurements is clinically important.
Purpose of the Study:
- To compare the inter- and intraobserver variability, accuracy, and acquisition time of tracked 3D ultrasound with automatic deep neural network segmentation against conventional 2D ultrasound for thyroid volumetry.
- To evaluate the performance of a convolutional neural network (CNN) for automatic thyroid lobe segmentation on 3D ultrasound data.
- To establish MRI as the reference standard for thyroid volume measurement in this comparison.
Main Methods:
- 28 healthy volunteers underwent thyroid scanning using both 2D and tracked 3D ultrasound by three physicians with varying experience levels.
- A convolutional deep neural network (CNN) was developed and trained for automatic segmentation of thyroid lobes on 3D ultrasound images.
- Thyroid volumes were measured using the ellipsoid formula for 2D ultrasound and automatic segmentation for 3D ultrasound, with MRI serving as the reference standard.
Main Results:
- Tracked 3D ultrasound with CNN segmentation demonstrated significantly reduced interobserver variability compared to 2D ultrasound.
- 3D ultrasound measurements showed no significant difference when compared to MRI reference volumes, unlike 2D ultrasound.
- Acquisition time for thyroid volumetry was significantly shorter using the 3D ultrasound technique.
Conclusions:
- Tracked 3D ultrasound combined with CNN-based automatic segmentation offers a more accurate and reproducible method for thyroid volumetry.
- This approach minimizes operator dependency, leading to more reliable thyroid volume measurements.
- The enhanced accuracy and efficiency of 3D ultrasound with AI segmentation hold promise for improved clinical management of thyroid diseases.
