Detection and Diagnostic Efficacy of Artificial Intelligence Ultrasound Assisted System for Thyroid Nodules Under Different Ultrasound Parameters / 中国医学影像学杂志

ABSTRACT

Purpose To explore the differences of the accuracy of detection and recognition of thyroid nodules and the diagnostic efficacy of benign and malignant thyroid nodules via artificial intelligence(AI)ultrasound assisted systems based on different ultrasound parameters.Materials and Methods A total of 147 patients with 289 nodules who underwent thyroid surgery in the First Medical Center of Chinese PLA General Hospital from March 30,2023 to May 1,2023 were prospectively selected.Different ultrasound parameters were adjusted and the AI system was used to detect and diagnose benign and malignant thyroid nodules via each parameter.Taken pathological results as the gold standard,the accuracy of thyroid nodule detection and the accuracy of benign and malignant diagnosis under different ultrasound parameters were compared,respectively.Results Under the standard ultrasound parameters,the accuracy of AI system in detecting thyroid nodules was 94.1%,the sensitivity for benign and malignant diagnosis was 90.9%,the specificity was 79.6%,and the accuracy was 86.6%,respectively.In terms of detection accuracy,accuracy under low gain(χ2=4.453,P=0.035)and high gain(χ2=6.215,P=0.013)parameters of AI system were significantly lower than those of standard ultrasound parameters.In terms of diagnostic efficacy,specificity(χ2=4.620,P=0.032),accuracy(χ2=7.521,P=0.006),area under the curve(Z=3.102,P=0.001),high gain sensitivity(χ2=6.170,P=0.013),accuracy(χ2=4.127,P=0.042),area under the curve(Z=2.152,P=0.031)and high depth accuracy(χ2=5.011,P=0.025),area under the curve(Z=2.420,P=0.015)of low gain were all significantly reduced compared to standard ultrasound parameters,with statistical differences.Conclusion When using the AI system to assist in the examination of thyroid nodules,attention should be paid to the adjustment of ultrasound instrument parameters.Improper parameter adjustment may reduce the AI system's ability to detect thyroid nodules and the accuracy of benign and malignant diagnosis.