Real-Time assessment of the beneficial role of artifical intelligence-based computer assisted diagnosis (AI-CAD) of thyroid nodules on ultrasound

Youngsook Kim; Miribi Rho; Sungjae Shin; Eunjung Lee; Daham Kim; Kwak Jin Young

doi:10.1530/endoabs.92.PS1-05-07

PS1-05-07

Prev Next

Section Contents Cite

Endocrine Abstracts (2023) 92 PS1-05-07 | DOI: 10.1530/endoabs.92.PS1-05-07

ETA2023 Poster Presentations Thyroid hormone diagnostics 1 (9 abstracts)

Real-Time assessment of the beneficial role of artifical intelligence-based computer assisted diagnosis (AI-CAD) of thyroid nodules on ultrasound

Youngsook Kim ¹ , Miribi Rho ² , Sungjae Shin ³ , Eunjung Lee ⁴ , Daham Kim ¹ & Jin Young Kwak ²

Err

Author affiliations

¹Severance Hospital, Institute of Endocrine Research, Yonsei University College of Medicine, Department of Internal Medicine, Seoul, Korea, Rep. of South; ²Severance Hospital, Research Institute of Radiological Science, Yonsei University College of Medicine, Department of Radiology, Seoul, Korea, Rep. of South; ³National Health Insurance Service Ilsan Hospital, Department of Internal Medicine, Goyang, Korea, Rep. of South; ⁴Yonsei University, School of Mathematics and Computing (Computational Science and Engineering), Seoul, Korea, Rep. of South

Objective: The purpose of this study is to evaluate and compare the effectiveness of artificial intelligence-based computer-assisted diagnosis (AI-CAD) in diagnosing thyroid malignancy by inexperienced physicians and experienced radiologist.

Methods: A total of 201 thyroid nodules from 192 patients was simultaneously evaluated by physician and radiologist using real-time ultrasound. After implementing AI-CAD, they reassessed the thyroid nodules. If necessary, the diagnosis was changed by referring to the AI-CAD results. Diagnostic performances of them with or without AI-CAD were calculated and compared.

Results: The sensitivity, specificity, diagnostic accuracy, positive predictive value (PPV), negative predictive value (NPV), area under the receiver operating characteristics (AUROC) was analyzed with/without AI-CAD assistance. Without AI-CAD, the area under the receiver operating characteristics curve (AUROC) of the radiologist (0.799) was higher than that of the inexperienced physician (0.704). With the assistance of AI-CAD, the AUROC increased to 0.814 for the radiologist and 0.729 for the inexperienced physician. Both of radiologist and inexperienced physician showed increased sensitivity (70.37% vs 73.33%, 75.56% vs 80.74%), diagnostic accuracy (76.62% to 78.61% vs 72.64% to 75.62%), PPV (93.14% to 93.40% vs 81.60% to 82.58%), NPV (59.60% to 62.11% vs 56.58% to 62.32%) with aid of AI-CAD, while specificity remained unchanged (89.39% vs 65.15%).

Conclusion: The diagnostic performance in differentiating thyroid nodules can be further improved with the assistance of AI-CAD, regardless of the level of experience, particularly for inexperienced physicians.