Tianjin Medical Journal ›› 2025, Vol. 53 ›› Issue (11): 1158-1164.doi: 10.11958/20252455

• Clinical Research • Previous Articles     Next Articles

Diagnostic efficacy of ultrasonic artificial intelligence combined with BRAF V600E gene testing in differentiating benign-malignant and invasive thyroid nodules

WU Menglin(), MA Fang, YANG Yafei   

  1. Department of Ultrasound, Hefei Second People's Hospital, Hefei 230000, China
  • Received:2025-07-10 Revised:2025-08-13 Published:2025-11-15 Online:2025-11-19

Abstract:

Objective To investigate the application value of ultrasonic artificial intelligence (AI) combined with serine/threonine-protein kinase (BRAF) V600E gene testing in differentiating benign-malignant and invasive thyroid nodules. Methods A total of 150 patients with malignant thyroid nodules (the malignant group) and 150 patients with benign thyroid nodules (the benign group) were selected. According to whether the pathological diagnosis of the malignant group involved capsule, vascular, nerve invasion or lymph node metastasis, patients were divided into the invasive group (66 cases) and the non-invasive group (84 cases). General clinical characteristics, ultrasonic AI parameters and BRAF V600E gene testing results were compared between groups. Discrepancies between ultrasonic AI, BRAF V600E gene testing and postoperative pathological diagnoses were analyzed. ROC curves and Delong tests were usd to evaluate the diagnostic efficacy of ultrasonic AI, BRAF V600E gene and their joint inspection. Results The malignant group exhibited higher probabilities of nodule maximum diameter (>1 cm), solid structure, hypoechoic/very hypoechoic echogenicity, microcalcification, blurred margin, irregular shape, aspect ratio (>1), internal and mixed blood flow distribution and high blood flow richness (grades Ⅲ—Ⅴ) compared to those of the benign group (P<0.05). The invasive subgroup showed higher probabilities of nodule maximum diameter (>1 cm), solid structure, hypoechoic/very hypoechoic echogenicity, microcalcification, blurred margin, irregular shape, internal and mixed blood flow distribution, and high blood flow richness (grades Ⅲ—Ⅴ) than those of the non-invasive subgroup (P<0.05). For diagnosing malignant thyroid nodules, ultrasonic AI demonstrated a sensitivity of 90.00% and specificity of 80.67%. For invasive malignant nodules, sensitivity was 84.85% and specificity was 83.33%. BRAF V600E gene testing showed a sensitivity of 72.67%, specificity of 90.00% for malignant nodules. For invasive nodules, sensitivity was 74.24% and specificity was 88.10%. Receiver operating characteristic curve (ROC) analysis revealed that the AUCs (95% CI) for ultrasonic AI, BRAF V600E gene and their joint inspection in diagnosing malignant thyroid nodules were 0.853 (0.807-0.900), 0.813 (0.762-0.864) and 0.941 (0.917-0.966), with the joint inspection outperforming individual tests (P<0.05). For invasive malignant nodules, the AUCs were 0.841 (0.773-0.909), 0.812 (0.737-0.886) and 0.924 (0.880-0.967), respectively, with the joint inspection showing superior performance (P<0.05). Conclusion The joint inspection ultrasound AI with BRAF V600E gene significantly improve the diagnostic efficacy for differentiating benign and malignant thyroid nodules, and assessing their invasive potential.

Key words: thyroid nodule, artificial intelligence, ultrasonography, BRAF V600E, benign-malignant, invasiveness

CLC Number: