探讨人工智能（AI）系统在乳腺影像报告与数据系统（BI-RADS）4类乳腺肿块良恶性鉴别诊断中的价值。

回顾性选取2018年1月至2020年2月于青岛市市立医院超声科初诊为BI-RADS 4类乳腺肿块的女性患者226例。所有患者均行常规超声检查，并经手术或穿刺活检取得病理结果。AI系统与不同年资乳腺超声专科医师（2、4、6年）分别对乳腺肿块超声图像进行分析并判断良恶性，应用四格表计算AI系统及不同年资医师对乳腺癌的诊断准确性，采用χ²检验比较AI系统与不同年资医师对不同大小乳腺癌肿块的诊断准确性。

226例乳腺肿块均经病理证实，其中良性病灶96例，恶性病灶130例。AI系统诊断乳腺恶性肿块的敏感度、特异度、阳性预测值、阴性预测值和准确性分别为93.84%、92.71%、94.57%、91.75%、93.36%，均高于不同年资医师。AI系统与不同年资医师诊断≤0.5 cm、＞0.5~1.0 cm、＞1.0~1.5 cm的乳腺癌肿块，其诊断准确性差异均有统计学意义（P=0.029、0.011、0.002）；诊断＞1.5~2.0 cm、＞2.0 cm的乳腺癌肿块，其诊断准确性差异均无统计学意义（P=0.117、0.668）。AI系统与2年资医师诊断≤0.5 cm、＞0.5~1.0 cm、＞1.0~1.5 cm的乳腺癌肿块，其诊断准确性差异均有统计学意义（P=0.006、0.002、0.001）。

AI系统在BI-RADS 4类乳腺肿块良恶性判断中具有较高的诊断价值，尤其对直径≤1.5 cm的乳腺癌的诊断；其可辅助低年资超声医师提高乳腺癌的诊断率。

To assess the diagnostic value of artificial intelligence(AI) system in the differential diagnosis of benign and malignant breast tumors of breast imaging reporting and data system(BI-RADS) grade 4.

A retrospective study was performed on 226 female patients with BI-RADS grade 4 breast masses from January 2018 to February 2020 at the Ultrasound Department of Qingdao Municipal Hospital. All the tumors were examined by routine ultrasonography and pathological results were obtained by operation or puncture biopsy. The AI system and breast ultrasound specialists with different years of experiences (2, 4, and 6 years) were used to analyze the breast mass ultrasound images and judge the lesion nature, and the diagnostic accuracy of the AI system and the doctors with different years of experience for breast cancer were calculated by the four-fold table method, and the χ² test was used to compare the diagnostic accuracy of AI system with that of physicians with different years of experience in breast cancer masses of different sizes.

A total of 226 breast masses were confirmed by pathology, including 96 benign lesions and 130 malignant lesions. The sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of the AI system were 93.84%, 92.71%, 94.57%, 91.75%, and 93.36%, respectively, which were higher than those of doctors with different years of experience. There were significant differences in diagnostic accuracy between the AI system and physicians with different years of experiences (P=0.029, 0.011, and 0.002, respectively) in breast cancer masses ≤0.5 cm,＞0.5-1.0 cm, and＞1.0-1.5 cm, although there was no significant difference in diagnostic accuracy of breast cancer masses＞1.5-2.0 cm and＞2.0 cm (P=0.117 and 0.668, respectively). There were significant differences in diagnostic accuracy between the AI system and physicians with 2 years of experience (P=0.006, 0.002, and 0.001, respectively) in breast cancer masses ≤0.5 cm,＞0.5-1.0 cm, and＞1.0-1.5 cm.

The AI system has high diagnostic value in the differentiation of benign and malignant breast masses of BI-RADS grade 4, especially in the diagnosis of breast cancer whose diameter is less than 1.5 cm, and it can assist junior doctors to improve the diagnostic rate of breast cancer.