Evaluation Of Artificial Intelligence(AI) Technology In Ultrasonic Diagnosis Of Thyroid Nodules

Objective:To explore the value of artificial intelligence（AI）technology in ultrasonic diagnosis of thyroid nodules.Methods:The data of 931 patients（1482 nodules）confirmed by operation and pathology in the Department of Ultrasound of the affiliated Union Hospital of Fujian Medical University from January 2018 to December 2018 were collected and analyzed by AI assistant diagnosis system of thyroid nodules.Taking the pathological results as the"gold standard",the ROC curves of Am CAD FJUN TI-RADS classification and Am CAD ACR TI-RADS classification of Am CAD-UT Detection system,AI-SONIC TI-RADS classification of AI-SONIC^TMThyroid system and FJUN TI-RADS classification examined by physicians were drawn respectively.The differences of AUC between groups were compared by Z test,the diagnostic effect was evaluated.The sensitivity,specificity and accuracy of the four methods were calculated,and the consistency between the results of the four methods and the pathological results were evaluated by Kappa Consistency test.According to the maximum diameter of thyroid nodules,patients were divided into 4 groups:micro-nodules（≤5mm）,small nodules（6mm～10mm）,medium-sized nodules（11mm～20mm）,large nodules（>20mm）.Four methods were used to retrospectively analyze the four groups of nodules,the diagnostic effects of four methods on thyroid nodules with different maximum diameter were analyzed.Depending on the quality of ultrasound images,nodules were divided into three groups:high image quality,medium image quality and general image quality.Four methods were used to retrospectively analyze the three groups of nodules,and the diagnostic efficiency of four methods for nodules with different image quality was analyzed.According to the qualifications of examiners,they were divided into three groups:high seniority,middle seniority and low seniority.Four methods were used to retrospectively analyze the three groups of nodules,and to analyze the difference of diagnostic accuracy of three methods of AI assistant diagnosis system in different seniority groups.Results:（1）The sensitivity of FJUN TI-RADS classification of physician was 86.58%,the specificity was 93.28%,and the diagnostic accuracy was 89.27%.The diagnostic efficiency of medium-sized nodules was the highest.（2）The sensitivity,specificity and accuracy of Am CAD FJUN TI-RADS classification system were 62.16%,79.97%and 69.30%,respectively.Among them,the Micro-nodules and high image quality have the highest specificity;（3）The sensitivity,specificity and accuracy of Am CAD ACR TI-RADS classification system were 96.28%,31.31%and70.24%,respectively.Among them,the nodules with the largest diameter and high image quality small nodules have the highest diagnostic efficiency;（4）The sensitivity of AI-SONIC TI-RADS classification diagnosis was 91.89%,the specificity was 81.82%,and the accuracy was89.85%.Among them,the nodules with the largest diameter and high image quality medium-sized nodules have the highest specificity;（5）The accuracy of AI-SONIC TI-RADS classification for medium-sized nodules was the highest,Am CAD ACR TI-RADS classification was the most sensitive for nodules with maximum diameter≤20mm.（6）Comparison of diagnostic efficacy of benign and malignant thyroid nodules:Physician FJUN TI-RADS classification was superior to AI-SONIC^TMThyroid system,and the latter was superior to Am CAD-UT Detection system.Conclusion:（1）At present,the diagnostic efficiency of AI assistant diagnosis system for thyroid nodules is higher,among which AI-SONIC^TMThyroid system is better than Am CAD-UT Detection system,but it is not as good as the examination level of doctors.（2）ACR TI-RADS classification is more suitable for Am CAD-UT Detection system.FJUN TI-RADS classification is more suitable for physician examination.（3）The size of thyroid nodules can affect the diagnostic efficiency of AI assistant diagnosis system.Am CAD ACR TI-RADS classification has higher sensitivity for nodules with maximum diameter≤20mm;Am CAD FJUN TI-RADS classification has higher specificity for micro-nodules,while AI-SONIC TI-RADS classification has higher sensitivity for medium-sized nodules.（4）The difference of image quality can affect the diagnostic efficiency of AI assistant diagnosis system for thyroid nodules.