OBJECTIVE: Ultrasound is a safe and timely diagnosis method commonly used for breast lesion, however it depends on the operator to a certain degree. As an emerging technology, artificial intelligence can be combined with ultrasound in depth to improve the intelligence and precision of ultrasound diagnosis and avoid diagnostic errors caused by subjectivity of radiologists. This paper aims to investigate the value of artificial intelligence S-detect system combined with virtual touch imaging quantification (VTIQ) technique in the differential diagnosis of benign and malignant breast masses by conventional ultrasound (CUS). respectively, and AUCs for them were 0.74, 0.86, 0.79, and 0.94, respectively. The diagnostic accuracy of CUS+S-detect was higher than that of CUS (P<0.05). The diagnostic accuracy of CUS+S-detect was higher than that of CUS (P<0.05). The diagnostic specificity of CUS+VTIQ was higher than that of CUS (P<0.05). The diagnostic accuracy and AUC of CUS+S-detect+VTIQ were higher than those of S-detect or VTIQ applied to CUS alone (P<0.05). The sensitivities of CUS for senior radiologists, CUS for junior radiologists, CUS+S-detect+VTIQ for senior radiologists, and CUS+S-detect+VTIQ for junior radiologists were 60.00%, 80.00%, 72.73%, and 90.00%, respectively, when diagnosing benign and malignant breast masses in 50 randomly selected cases. The specificities for them was 66.67%, 76.67%, 80.00%, and 81.25%, respectively. The accuracies for them was 64.00%, 78.00%, 80.00%, and 88.00%, respectively. The AUCs for them were 0.63, 0.78, 0.88, and 0.80, respectively. Compared with the CUS for junior radiologists, the CUS+S-detect+VTIQ for junior radiologists had higher sensitivity, specificity, and accuracy (all P<0.05). The consistency of the combined application of S-detect and VTIQ for diagnosing breast masses at 2 different times was good among junior radiologists (Kappa=0.800).
METHODS: CUS, S-detects, and VTIQ were used to differentially diagnose benign and malignant breast masses in 108 cases, and the final pathological results were referred to as the gold standard for classifying breast masses. The diagnostic efficacy were evaluated and compared, among the 3 methods and among S-detect applied to CUS (CUS+S-detect), VTIQ applied to CUS (CUS+VTIQ), and S-detect combined with VTIQ applied to CUS (CUS+S-detect+VTIQ). Fifty cases were acquired randomly from the collected breast masses, and 2 radiologists with different years of experience (2 and 8 years) used S-detect combined with VTIQ for the ultrasonic differential diagnosis of benign and malignant breast masses.
RESULTS: The differences in sensitivity, specificity, accuracy, and the area under the receiver operating characteristic curve (AUC) of the 3 diagnostic methods of CUS, S-detect, and VTIQ were not statistically significant (all P>0.05). The sensitivities of CUS, CUS+Sdetect, CUS+VTIQ, and CUS+S-detect+VTIQ were 78.57%, 92.86%, 69.05%, and 95.24%, respectively, the specificities for them were 69.70%, 78.79%, 87.88%, and 92.42%, respectively, the accuracies for them were 73.15%, 84.26%, 80.56%, and 93.52%.
CONCLUSIONS: S-detect combined with VTIQ when applied to CUS can overcome the shortcomings of separate applications and complement each other, especially for junior radiologists, and can more effectively improve the diagnostic efficacy of ultrasound for benign and malignant breast masses.
目的: 超声作为乳腺病变诊断的常用方法,具有实时、安全等优势,但存在一定的操作者依赖性。人工智能是近年来发展的新技术,与超声深度结合有助于提高超声诊断的智能化和精准化,避免出现超声医师因主观因素带来的诊断误差。本研究旨在探讨人工智能S-detect系统联合声触诊组织成像定量(virtual touch imaging quantification,VTIQ)技术在常规超声(conventional ultrasound,CUS)鉴别诊断乳腺肿块良恶性中的应用价值。方法: 采用CUS、S-detect及VTIQ对108个乳腺肿块进行良恶性鉴别诊断,以病理结果为金标准,比较3种诊断方法的诊断效能,并对比分析S-detect应用于CUS(CUS+S-detect)、VTIQ应用于CUS(CUS+VTIQ)以及S-detect联合VTIQ应用于CUS(CUS+S-detect+VTIQ)的诊断效能。从收集的乳腺肿块图像中随机抽取50个,由2名不同年资的医师(分别有2年及8年经验),分别应用S-detect联合VTIQ,对乳腺肿块进行良恶性鉴别诊断,比较不同年资医师应用S-detect联合VTIQ之间以及同一医师内部诊断的一致性。结果: CUS、S-detect及VTIQ 3种诊断方法的灵敏度、特异度、准确度、AUC比较差异均无统计学意义(均P>0.05)。CUS、CUS+S-detect、CUS+VTIQ、CUS+S-detect+VTIQ的灵敏度分别为78.57%、92.86%、69.05%、95.24%;特异度分别为69.70%、78.79%、87.88%、92.42%;准确度分别为73.15%、84.26%、80.56%、93.52%;受试者工作特征(receiver operating characteristic,ROC)曲线下面积(area under the curve,AUC)分别为0.74、0.86、0.79、0.94。CUS+S-detect的诊断准确度高于CUS(P<0.05);CUS+VTIQ的诊断特异度高于CUS(P<0.05);CUS+S-detect+VTIQ三者联合应用的诊断准确度和AUC均高于S-detect或VTIQ单独应用于CUS(均P<0.05)。高年资医师CUS、低年资医师CUS、高年资医师CUS+S-detect+VTIQ、低年资医师CUS+S-detect+VTIQ分别诊断随机抽选的50个乳腺肿块的良恶性时,其灵敏度分别为60.00%、80.00%、72.73%、90.00%;特异度分别为66.67%、76.67%、80.00%、81.25%;准确度分别为64.00%、78.00%、80.00%、88.00%;AUC分别为0.63、0.78、0.88、0.80。与低年资医师CUS相比,低年资医师CUS+S-detect+VTIQ具有更高的灵敏度、特异度和准确度(均P<0.05)。低年资医师在2次不同时间联合应用S-detect和VTIQ于CUS诊断乳腺肿块的一致性较好(Kappa=0.800)。结论: S-detect联合VTIQ应用于CUS时,可克服单独应用的不足,尤其能有效提高低年资医师对乳腺良恶性肿块的诊断效能。.