S-Detect技术在乳腺癌鉴别诊断中的辅助诊断价值

The assistant diagnostic value of S-Detect technique in identification of breast cancer

目的评估S-Detect人工智能系统在乳腺良恶性肿块鉴别中的辅助诊断价值。方法采集2018年3-5月于华中科技大学同济医学院同济医院超声影像科进行乳腺超声检查的201例患者的超声图像,分析220个乳腺病灶的常规超声图像、S-Detect模式图像及弹性图像。将每个乳腺肿块的BI-RADS分类结果分为高年资医师组与随机组。根据是否借助S-Detect诊断结果,将高年资组的BI-RADS分类分为A1组与P1组。将随机组中同一病灶的最高、最低分类两次结果归为A2组,并将A2组与S-Detect系统联合诊断结果记为P2组。绘制ROC曲线,比较不同组诊断结果的ROC曲线下面积、敏感性、特异性和准确性。用Kappa检验分析不同方式诊断的一致性。结果220个病灶中良性181个,恶性39个。S-Detect人工智能系统有着较高的诊断效能,其敏感性、特异性与准确性分别为92.3%、90.6%、90.9%。在该技术辅助下,高年资超声医师诊断的特异性、准确性有上升趋势(A1组:86.7%、88.6%;P1组:91.2%、92.3%),医师整体诊断的准确性显著提高(A2组:63.6%~85.5%;P2组:93.2%~94.1%)。S-Detect技术与弹性评分技术均能提升超声医师对乳腺癌诊断的效能,但是二者的诊断能力与辅助诊断能力有所差异。结论S-Detect人工智能系统有助于提升超声医师诊断乳腺癌的准确性,提高患者所获随机乳腺超声检查的质量,减少漏诊、误诊。

Objective To evaluate the assistant diagnostic value of S-Detect artificial intelligence system in differential diagnosis of benign and malignant breast tumors. Methods Clinical data and ultrasound images of 201 patients undergoing breast ultrasound examination in Tongji Hospital from March 2018 to May 2018 were acquired. Two-dimensional grayscale and color Doppler ultrasound images, S-Detect mode images and elastographic images of 220 breast lesions were analyzed. The BI-RADS categories of each lesion were divided into two groups: experienced group and random group.And according to whether to refer to S-Detect diagnostic results, the BI-RADS categories in experienced group were divided into A1 group and P1 group.In additional, the highest and lowest categories of the same tumor in random group were A2 group, and the diagnostic results of A2 group combining with S-Detect system were belonged to P2 group. The ROC curves were plotted and the area under the curve, sensitivity, specificity or the accuracy of the different groups were compared. Agreements of diagnostic results between different groups were analyzed by Kappa test. Results Out of 220 breast lesions, 181 lesions were benign and 39 lesions were malignant. The S-Detect artificial intelligence system had a relatively high diagnostic efficiency, and the sensitivity, specificity and accuracy of S-Detect classification were 92.3%, 90.6%, 90.9%, respectively. With its assistance, the specificity and accuracy in the experienced group had an increasing trend (A1 group: 86.7%, 88.6%;P1 group: 91.2%, 92.3%), and the diagnostic accuracy in random group was significantly improved (A2 group: 63.6%-85.5%;P2 group: 93.2%-94.1%). Both S-Detect system and elasticity score helped to improve the efficacy of ultrasound physicians in differential diagnosis of benign and malignant breast lesions. But there were differences in diagnostic performance and assistant diagnostic ability between the two techniques. Conclusions S-Detect technique contributes to the augment of diagnostic accuracy of ultrasound doctors in identifying breast cancer, improves the quality of random breast ultrasound examinations, and reduces missed diagnosis and misdiagnosis of breast examinations.