Diagnostic Performance of Artificial Intelligence for Interpreting Thyroid Cancer in Ultrasound images

Piyanuch Arunrukthavon, Dittapong Songsaeng, Chadaporn Keatmanee, Songphon Klabwong, Mongkol Ekpanyapong, Matthew N. Dailey

Source Title: International Journal of Knowledge and Systems Science (IJKSS)13(1)

ISSN: 1947-8208|EISSN: 1947-8216|EISBN13: 9781683181903|DOI: 10.4018/IJKSS.309431

Cite Article Cite Article

MLA

Arunrukthavon, Piyanuch, et al. "Diagnostic Performance of Artificial Intelligence for Interpreting Thyroid Cancer in Ultrasound images." IJKSS vol.13, no.1 2022: pp.1-13. http://doi.org/10.4018/IJKSS.309431

APA

Arunrukthavon, P., Songsaeng, D., Keatmanee, C., Klabwong, S., Ekpanyapong, M., & Dailey, M. N. (2022). Diagnostic Performance of Artificial Intelligence for Interpreting Thyroid Cancer in Ultrasound images. International Journal of Knowledge and Systems Science (IJKSS), 13(1), 1-13. http://doi.org/10.4018/IJKSS.309431

Chicago

Arunrukthavon, Piyanuch, et al. "Diagnostic Performance of Artificial Intelligence for Interpreting Thyroid Cancer in Ultrasound images," International Journal of Knowledge and Systems Science (IJKSS) 13, no.1: 1-13. http://doi.org/10.4018/IJKSS.309431

Export Reference

Favorite Full-Issue Download

View Full Text HTML

View Full Text PDF

Abstract

Thyroid ultrasonography is mainly used for the detection and characterization of thyroid nodules. However, there is some limitation since the diagnostic performance remains highly subjective and depends on radiologist experiences. Therefore, artificial intelligence (AI) was expected to improve the diagnostic performance of thyroid ultrasound. To evaluate the diagnostic performance of the AI for differentiating malignant and benign thyroid nodules and compare it with that of an experienced radiologist and a third-year diagnostic radiology resident, 648 patients with 650 thyroid nodules, who underwent thyroid ultrasound guided-FNA biopsy and had a decisive diagnosis from FNA cytology at Siriraj Hospital between January 2014 and June 2020, were enrolled. Although the specificity and accuracy were slightly higher in AI than the experienced radiologist and the resident (specificity 78.85% vs. 67.31% vs. 69.23%; accuracy 78.46% vs. 70.77% vs. 70.77%, respectively), the AI showed comparable diagnostic sensitivity and specificity to the experienced radiologist and the resident (p=0.187-0.855).