Related Experiment Video
Updated: Jul 19, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Improving the Efficacy of ACR TI-RADS Through Deep Learning-Based Descriptor Augmentation
Lev Barinov1,2, Ajit Jairaj3, William D Middleton4
1Department of Radiology, University of Pennsylvania, Philadelphia, PA, USA. Lev.Barinov@pennmedicine.upenn.edu.
Journal of Digital Imaging
|August 14, 2023
Summary
Artificial intelligence decision support systems significantly enhance physician accuracy in diagnosing thyroid nodules. Integrating AI tools like Koios DS improves diagnostic metrics and reduces interpretation time, aiding clinical decisions.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Healthcare
- Diagnostic Decision Support
Background:
- Thyroid nodules are common, with a small percentage being malignant, yet many fine-needle aspirations (FNAs) are performed annually.
- Current diagnostic methods for thyroid nodules have limitations in accuracy and efficiency.
- Artificial intelligence (AI) offers potential to improve diagnostic accuracy and workflow in clinical pathways.
Purpose of the Study:
- To evaluate the impact of an AI-based decision support system (DSS), Koios DS, on physician diagnostic accuracy, variability, and efficiency for thyroid nodules.
- To compare physician performance with and without AI assistance in assessing thyroid nodule malignancy risk.
- To analyze the system's alignment with the American College of Radiology Thyroid Imaging Reporting and Data System (ACR TI-RADS).
Main Methods:
- Retrospective analysis of 650 patient cases (thyroid nodules) between August and October 2020.
- Fifteen physicians assessed cases both with and without the Koios DS decision support system.
- Diagnostic accuracy measured by Area Under the ROC Curve (AUC), sensitivity, and specificity, with FNA as ground truth.
- Reader variability assessed using Pearson's correlation coefficient, and efficiency by average interpretation time per case.
Main Results:
- AI-assisted physician performance showed a statistically significant increase in average AUC (0.083), sensitivity (8.4%), and specificity (14%).
- The average interpretation time per case decreased by 23.6% with AI assistance.
- Inter-reader variability significantly reduced, with Pearson's correlation coefficient increasing from 0.622 to 0.876 when using the DSS.
Conclusions:
- AI-driven decision support systems, such as Koios DS, significantly enhance physician diagnostic accuracy for thyroid nodules.
- The integration of AI tools leads to reduced inter-reader variability and improved interpretation efficiency.
- Automated clinical decision support has a substantial positive impact on diagnostic pathways for thyroid nodules.
