Diagnostic performance and clinical integration of an artificial intelligence–assisted ultrasound decision support tool for thyroid nodules at a tertiary cancer centre

Purpose: To evaluate the diagnostic performance of radiologists, with and without an artificial intelligence tool, at a tertiary cancer centre. Methods: The institutional review board considered this study at a tertiary cancer centre as a quality improvement project. Pre-biopsy ultrasound images of thyroid nodules obtained by a single interventional radiologist between July 2023 and April 2025 were retrospectively evaluated by five subspecialised diagnostic radiologists independently for thyroid imaging reporting and data system assessment, first without and then with the artificial intelligence tool. Standalone artificial intelligence assessments were also recorded. Diagnostic performance metrics were assessed, including conditional sensitivity, conditional specificity and conditional area under the curve. Results: Ninety nodules from 73 patients (50/73, 68%, female; mean age, 56 ± 17 years) were included. Pathology and genetic testing classified 46/90 nodules (51%) as benign, 31/90 (34%) as malignant and 13/90 (14%) as indeterminate. Compared to radiologist interpretation alone, artificial intelligence assistance yielded slight improvements in conditional specificity (44% [95% confidence interval: 22%, 69%] vs 38% [95% confidence interval: 15%, 65%]), high risk predictive value (58% [95% confidence interval: 37%, 78%] vs 57% [95% confidence interval: 35%, 77%]), area under the receiver operating characteristic curve (0.78 vs 0.75), overall test yield (42% [95% confidence interval: 30%, 53%] vs 38% [95% confidence interval: 27%, 49%]) in classifying benign and malignant nodules, though not statistically significant. Conclusion: Combined radiologist and artificial intelligence tool interpretations achieved similar diagnostic performance to interpretation by radiologists alone or Koios DS Thyroid alone, demonstrating utility of artificial intelligence–assisted interpretation in a specialised cancer centre with a high prevalence of rare and aggressive thyroid and non-thyroid malignancies.

Authors

Institutions

Publication Details

Journal
Ultrasound
Published
2026-11-16
DOI
https://doi.org/10.1177/1742271x261503106
Primary Topic
Thyroid Cancer Diagnosis and Treatment
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Diagnostic performance and clinical integration of an artificial intelligence–assisted ultrasound decision support tool for thyroid nodules at a tertiary cancer centre

Wyanne Law, Marinela Capanu, Shannan M. Dickinson, Jennifer S. Golia Pernicka et al.
3 links
Ultrasound
Thyroid Cancer Diagnosis and Treatment
article

Diagnostic performance and clinical integration of an artificial intelligence–assisted ultrasound decision support tool for thyroid nodules at a tertiary cancer centre

Wyanne Law, Marinela Capanu, Shannan M. Dickinson, Jennifer S. Golia Pernicka, Duan Li, Junting Zheng, Pamela Argiriadi, Debra Sarasohn, Ariadne Bach
article en

Abstract

Purpose: To evaluate the diagnostic performance of radiologists, with and without an artificial intelligence tool, at a tertiary cancer centre. Methods: The institutional review board considered this study at a tertiary cancer centre as a quality improvement project. Pre-biopsy ultrasound images of thyroid nodules obtained by a single interventional radiologist between July 2023 and April 2025 were retrospectively evaluated by five subspecialised diagnostic radiologists independently for thyroid imaging reporting and data system assessment, first without and then with the artificial intelligence tool. Standalone artificial intelligence assessments were also recorded. Diagnostic performance metrics were assessed, including conditional sensitivity, conditional specificity and conditional area under the curve. Results: Ninety nodules from 73 patients (50/73, 68%, female; mean age, 56 ± 17 years) were included. Pathology and genetic testing classified 46/90 nodules (51%) as benign, 31/90 (34%) as malignant and 13/90 (14%) as indeterminate. Compared to radiologist interpretation alone, artificial intelligence assistance yielded slight improvements in conditional specificity (44% [95% confidence interval: 22%, 69%] vs 38% [95% confidence interval: 15%, 65%]), high risk predictive value (58% [95% confidence interval: 37%, 78%] vs 57% [95% confidence interval: 35%, 77%]), area under the receiver operating characteristic curve (0.78 vs 0.75), overall test yield (42% [95% confidence interval: 30%, 53%] vs 38% [95% confidence interval: 27%, 49%]) in classifying benign and malignant nodules, though not statistically significant. Conclusion: Combined radiologist and artificial intelligence tool interpretations achieved similar diagnostic performance to interpretation by radiologists alone or Koios DS Thyroid alone, demonstrating utility of artificial intelligence–assisted interpretation in a specialised cancer centre with a high prevalence of rare and aggressive thyroid and non-thyroid malignancies.

Ultrasound
Memorial Sloan Kettering Cancer Center (US)
Good health and well-being
Thyroid Cancer Diagnosis and Treatment
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon