Study On The Clinical Application Of AI Combined With CEUS In Partially Cystic Thyroid Carcinoma

Objectives:Preliminary explore the contrast-enhanced ultrasound(CEUS)enhancement mode of partial cystic thyroid nodules(PCTNs),and evaluate its diagnostic efficacy for PCTNs by establishing the CEUS enhancement mode scoring standard;To study the diagnostic efficacy of AI(Artificial Intelligence,AI)software based on ACR TI-RADS classification for PCTNs,and to explore the diagnostic efficacy of AI software for assisting physicians using ACR TIRADS classification combined with CEUS for PCTNs.Methods:A total of 106 patients with PCTNs in our hospital from March 2013 to December 2020 were collected,with a total of 115 nodules.All patients were confirmed pathologically by surgery or biopsy.The patients were divided into benign and malignant groups based on the pathological results.All patients underwent routine ultrasound and CEUS examinations before surgery.1.The CEUS enhancement mode of PCTNs was summarized by consulting related literature and observation,and the CEUS enhancement mode scoring standard was established to analyze its diagnostic efficacy for PCTNs.2.Test the diagnostic efficacy of AI software on PCTNs based on the ACR TI-RADS classification,and compare the diagnostic efficacy of the AI software before and after the application of the ACR TI-RADS classification to PCTNs by the AI software assisted physicians,and observe the combined diagnostic efficacy.3.Compared the scoring standard of CEUS enhanced mode with that of doctors before and after applying the ACR TI-RADS classification combined with CEUS enhanced mode scoring standard to observe the combined diagnostic efficacy of PCTNs.4.To explore the diagnostic efficacy of AI software assisted physicians using ACR TIRADS classification combined with CEUS enhanced model scoring criteria for PCTNs.Results:1.There are differences in the CEUS enhancement mode of benign and malignant PCTNs.Among them,the enhanced ring(P<0.001),island-like enhancement(P=0.016),whether the nodule boundary is clear after enhancement(P<0.001),the size of the solid part changes after enhancement(P < 0.001),sparseness is low/no enhancement(P <0.001),substantial part of enhancement time(P=0.010)and regression time(P=0.004)were different in the two groups.2.The AUC,sensitivity,specificity,accuracy,PPV,and NPV of the CEUS enhanced model score in the diagnosis of PCTNs were 0.849,75.4%,81.0%,78.3%,79.6%,and 77.0%,respectively.3.The AUC,sensitivity,specificity,accuracy,PPV,and NPV of the AI software in the diagnosis of PCTNs were 0.755,82.5%,67.2%,74.8%,71.2%,and 79.6%,respectively.4.The AUC,sensitivity,specificity,accuracy,PPV,and NPV of the physicians using ACR TI-RADS classification to diagnose PCTNs were 0.880,84.2%,82.8%,83.5%,82.8%,and 84.2%,respectively.5.The AUC,sensitivity,specificity,accuracy,PPV,and NPV of the AI software assisted physicians in the diagnosis of PCTNs were 0.929,91.2%,86.2%,88.7%,86.7%,and 90.9%,respectively.Comparing the diagnostic efficacy of AI software and physicians using the ACR TI-RADS classification for PCTNs,the sensitivity of AI software in diagnosing PCTNs is equivalent to that of physicians,and the difference is not statistically significant(P=1.000);the specificity and AUC of AI software in diagnosing PCTNs are lower than that of PCTNs.Doctors,the difference was statistically significant(P=0.012;P=0.002).After AI software assisted physicians,the physicians’ diagnostic indicators of PCTNs were improved to varying degrees.Compared with the specificity of AI software alone in diagnosing PCTNs,the difference was statistically significant(P=0.003);compared with the AUC of PCTNs diagnosed by AI software and physicians alone,the difference was statistically significant(P<0.001;P=0.045).6.The AUC,sensitivity,specificity,accuracy,PPV,and NPV of the physicians using ACR TI-RADS classification combined with CEUS to diagnose PCTNs were 0.914,91.2%,84.5%,88.7%,85.2%,and 90.7%,respectively.Comparing the diagnostic efficacy of CEUS and physicians using ACR TI-RADS classification for PCTNs,the two have the same diagnostic efficacy for PCTNs,and the difference is not statistically significant(both P>0.05).After physicians used the ACR TI-RADS classification combined with CEUS,the various indicators for the diagnosis of PCTNs were improved to varying degrees.Compared with the sensitivity of CEUS alone in diagnosing PCTNs,the difference was statistically significant(P=0.004);compared with the AUC diagnosed by physicians and CEUS alone,the difference was statistically significant(P=0.009;P=0.036).7.After the AIsoftware assisted physicians using the ACR TI-RADS classification combined with CEUS,various indicators for the diagnosis of PCTNs were improved to varying degrees.Among them,compared with the sensitivity of AI software and CEUS alone in the diagnosis of PCTNs,the differences were statistically significant(P=0.031;P=0.001);compared with the specificity of AI software alone in diagnosing PCTNs,the difference is statistically significant(P=0.002);compared with AI software,CEUS and the AUC of PCTNs diagnosed by physicians alone,the difference is statistically significant(P<0.001;P=0.004;P=0.006).Conclusions:1.There are differences in the CEUS enhancement mode of benign and malignant PCTNs.The CEUS scoring standard can effectively diagnose PCTNs,and the diagnostic efficiency is basically equivalent to that of physicians using ACR TI-RADS classification.Physicians’ application of ACR TI-RADS classification combined with CEUS can effectively improve the diagnostic performance of PCTNs.After the combination,the AUC of the two separate diagnosis is improved,and the sensitivity of the diagnosis of CEUS alone is improved.2.AI software can effectively diagnose PCTNs;after assisting physicians,AI software can effectively improve the diagnostic efficiency of PCTNs,and at the same time make up for the shortcomings of AI software’s lower specificity,avoid over-diagnosis,and compare AI software and physicians alone diagnosis after combination AUC is improved.3.Physicians using ACR TI-RADS classification can effectively diagnose PCTNs.AI software assists physicians using ACR TI-RADS classification combined with CEUS to have the best diagnostic efficiency.The combination has improved sensitivity,specificity,and AUC compared with AI software alone.Compared with CEUS alone,the sensitivity and AUC of diagnosis are improved,and compared with the AUC of physicians alone,the combination of the three is expected to achieve accurate diagnosis of PCTNs.