ECR 2026: Poster Presentation
Sadaf Monajemi, Leeann D. Louis, Annie Ng, Melih Engin, Shruti Surendrakumar, Sebastian Parkitny, Jamie Vo, Milad Mohammadzadeh, Bryan Haslam
Abstract
Purpose and Learning Objective
Thyroid nodule characterisation on ultrasound is time-intensive and prone to substantial inter-radiologist variability and errors due to the multi-step requirements of ACR TI-RADS (TR) assessment and structured reporting. We evaluate an AI-enhanced thyroid ultrasound assistant that fully automates nodule detection, characterisation, and report generation following TR requirements, and assess its impact on efficiency, agreement, and diagnostic accuracy.
Methods or Background
Interpretation time, TR-level agreement, and nodule characterisation accuracy were assessed in a blinded, fully crossed, multi-reader, multi-case study in which 18 radiologists (14 US-board certified, four Australia/New Zealand-board certified) interpreted 600 cases (74% from the US, 26% from Australia; 104 biopsy-confirmed malignant and 496 biopsy-confirmed or two-year follow-up confirmed negative) both unaided and aided by the AI assistant with a four-week washout period in between. Negative cases were randomly selected and enriched to ensure adequate representation across all TR levels. Ground truth for TR levels and descriptors was established by consensus by three radiologists. 95% confidence intervals (CI) were calculated using bootstrapping.
Results
Use of the AI assistant significantly (p<0.001) reduced interpretation time (44.8 [95% CI: 43.8-46.3] seconds aided vs 72.4 [70.8-74.0] unaided) and significantly increased inter-radiologist TR-level agreement (62.3% [59.1-65.5] vs 53.9% [50.6-57.4]). Accuracy improved across all TI-RADS descriptors: composition (84.9% vs 80.4%, difference 4.8% [95% CI: 2.8-6.7]); echogenicity (77.4% vs 70.0%, difference 7.5% [5.1-10.0]); shape (90.8% vs 86.4%, difference 4.4% [3.0-6.0]); margin (73.5% vs 57.3%, difference 16.2% [12.4-20.1]); and echogenic foci (75.2% vs 71.1%, difference 4.1% [2.0-6.4]).
Conclusion
An AI-enhanced thyroid ultrasound assistant can improve diagnostic efficiency, consistency, and accuracy and thus enhance overall quality of care.
Disclaimers:
Disclaimers: DeepHealth Thyroid Suite includes DeepHealth Viewer and DeepHealth Thyroid AI. DeepHealth Viewer is Manufactured as eRAD PACS by eRAD and distributed by DeepHealth. DeepHealth Thyroid AI is manufactured as See-Mode Augmented Reporting Tool, Thyroid (SMART-T) by See-Mode and distributed by DeepHealth Inc. Any claims made about Thyroid Suite may reference claims associated with its individual components. Not all products and functionalities are commercially available in all countries.
Any claims made about Thyroid Suite may reference claims associated with its individual components.
Not all products and functionalities are commercially available in all countries. For clearance and commercial availability in your geography of functionalities listed and compatibility with other systems, please contact your account manager.