Ultrasound-based computer-aided diagnosis for cytologically indeterminate thyroid nodules with different radiologists

Clin Hemorheol Microcirc. 2022;82(3):217-230. doi: 10.3233/CH-221423.

Abstract

Purpose: To evaluate a computer-aided diagnosis (CAD) technique in predicting malignancy for cytologically indeterminate thyroid nodules (TNs) as compared with different experienced radiologists.

Method: 436 patients with 436 cytologically indeterminate TNs on fine-needle aspiration cytology (FNAC) were included and all were confirmed by surgical pathology. They were retrospectively analyzed with respect to ultrasound (US) characteristics using a commercially available CAD system (AmCAD-UT; AmCad BioMed, Taiwan, China) and reviewed by one junior and one senior radiologists.The CAD system and different experienced radiologists stratified the risk of malignancy using ACR TI-RADS category. The diagnostic performance by different experienced radiologists independently and after consulting the CAD (different experienced radiologists + CAD) and by the CAD alone were compared.

Results: The different experienced radiologists showed significantly higher specificities than the CAD system alone. The combination of radiologist and CAD system showed improved diagnostic performance with an AUC (Area under the curve) of 0.740 in the senior radiologist and 0.677 in the junior radiologist, as compared with CAD (AUC: 0.585) alone (all P < 0.05). The combination of senior radiologist and CAD system had the highest diagnostic performance (AUC: 0.740) and specificity (68.9%) compared to the others (all P < 0.05).

Conclusion: The CAD system may play the potential role as a decision-making assistant alongside radiologists for differential diagnosis of TNs with indeterminate cytology.

Keywords: Computer-aided diagnosis; thyroid; ultrasound.

MeSH terms

  • Computers
  • Humans
  • Radiologists
  • Retrospective Studies
  • Sensitivity and Specificity
  • Thyroid Neoplasms* / pathology
  • Thyroid Nodule* / pathology
  • Ultrasonography / methods