基于人工智能深度学习算法的超声诊断系统在触诊阴性的乳腺结节良恶性鉴别中的应用  

Application of ultrasound diagnosis system based on artificial intelligence deep learning algorithm in identification of benign and malignant breast nodules with negative palpation

作　　者：刘瑞 袁文佳 刘巍 LIU Rui;YUAN Wenjia;LIU Wei(Department of Ultrasound,the Affiliated Cancer Hospital,Zhengzhou University(Henan Cancer Hospital),Zhengzhou 450003)

机构地区：[1]郑州大学附属肿瘤医院(河南省肿瘤医院)超声科,郑州450003

出　　处：《郑州大学学报（医学版）》2023年第3期406-410,共5页Journal of Zhengzhou University(Medical Sciences)

摘　　要：目的:探讨基于人工智能深度学习算法的超声诊断系统在触诊阴性的乳腺结节良恶性鉴别中的应用。方法:回顾性分析120例乳腺结节患者临床资料,所有纳入的病例术前触诊阴性,并行超声检查、超声弹性成像以及人工智能深度学习算法S-Detect检查,术后有完整的病理组织学报告。分别计算超声弹性成像、人工智能S-Detect以及联合诊断方法对于乳腺良恶性结节的诊断价值。结果:120例患者共检查出153个乳腺病灶,组织病理学检查中良性病灶共97个,恶性病灶共56个。超声弹性成像共检出60个恶性结节和93个良性结节,超声弹性成像诊断敏感度75.00%,特异度81.44%,阳性预测值70.00%,阴性预测值84.95%,准确度79.08%。人工智能S-Detect共检出67个恶性结节和86个良性结节,人工智能S-Detect诊断敏感度91.07%,特异度83.51%,阳性预测值76.12%,阴性预测值94.19%,准确度86.27%。二者联合诊断共检出65个恶性结节和88个良性结节,诊断敏感度94.64%,特异度87.63%,阳性预测值81.54%,阴性预测值96.59%,准确度90.20%。与单独超声弹性成像或人工智能S-Detect相比,联合诊断方案AUC(95%CI)为0.864(0.790~0.942),具有更好的诊断效果。结论:基于人工智能深度学习算法的超声诊断系统在触诊阴性的乳腺结节良恶性鉴别中的应用效果较好,有助于辅助临床诊断。Aim:To explore the application of ultrasonic diagnosis system based on artificial intelligence deep learning algorithm in the differentiation of benign and malignant breast nodules with negative palpation.Methods:Clinical data of 120 patients with breast nodules were retrospectively analyzed.All the included cases were negative in palpation before surgery.Ultrasonic examination,ultrasonic elastography and deep learning algorithm artificial intelligent S-Detect(S-Detect)were performed.A complete histopathological report was provided after surgery.The value of ultrasonic elastography,S-Detect and combination of the 2 methods for the diagnosis of benign and malignant breast nodules were calculated respectively.Results:A total of 153 breast lesions were detected in 120 patients.There were 97 benign lesions and 56 malignant lesions according to histopathological results.A total of 60 malignant nodules and 93 benign nodules were detected by ultrasonic elastography;the diagnostic sensitivity of ultrasonic elastography was 75.00%,the specificity was 81.44%,the positive predictive rate was 70.00%,the negative predictive value was 84.95%,and the accuracy was 79.08%.A total of 67 malignant nodules and 86 benign nodules were detected by S-Detect;the diagnostic sensitivity of S-Detect was 91.07%,the specificity was 83.51%,the positive predictive rate was 76.12%,the negative predictive value was 94.19%,and the accuracy was 86.27%.S-Detect combined with ultrasonic elastography detected 65 malignant nodules and 88 benign nodules;the diagnostic sensitivity was 94.64%,the specificity was 87.63%,the positive predictive rate was 81.54%,the negative predictive value was 96.59%,and the accuracy was 90.20%.The AUC(95%CI)of the joint diagnosis scheme was 0.864(0.790-0.942),which was better than that of the ultrasonic elastic imaging or S-Detect alone.Conclusion:The ultrasonic diagnosis system based on artificial intelligence deep learning algorithm has a good application effect in the differentiation of benign and malignant breast nodul

关 键 词：乳腺结节 良恶性鉴别 超声诊断 人工智能 

分 类 号：R737.9[医药卫生—肿瘤]