基于人工智能的软件在靶向超声发现的乳腺病灶中预测恶性风险的表现
回顾性横断面单中心研究纳入334个MRI发现后经靶向超声和超声引导穿刺活检的乳腺病灶,其中良性257个(76.9%)、恶性77个(23.1%)。AI软件与放射科医生均显示高敏感度,放射科医生结合AI软件评估的特异度高于单独放射科医生评估(p < 0.001)。放射科医生或AI软件判为BI-RADS 2或3类的72个(21.6%)病灶病理均为良性。
要点
- 放射科医生结合AI软件评估靶向超声病灶,特异度高于单独放射科医生评估(p < 0.001)。
- 放射科医生或AI软件判为BI-RADS 2或3类的72个(21.6%)病灶病理均为良性。
- AI软件与放射科医生对靶向超声病灶恶性风险预测均显示高敏感度。
核心结论
- AI软件整合到放射科医生评估中,对靶向超声的质量和非质量病灶均显示高诊断准确度并提高特异度,支持更准确的活检决策,并可能在不漏诊癌症的情况下减少假阳性。
研究设计与样本
- 回顾性横断面单中心研究,纳入MRI发现乳腺病灶并接受靶向超声及经皮超声引导活检的患者,共评估334个病灶(质量病灶183个、非质量病灶151个),超声表现用AI软件分析并与病理结果对照。
诊断标准与关键数值
- 病理结果:良性257个(76.9%),恶性77个(23.1%)。
- 放射科医生或AI软件判为BI-RADS 2或3类的病灶共72个(21.6%),病理均为良性。
临床应用与阅片要点
- 放射科医生结合AI软件评估靶向超声病灶,可提高特异度(p < 0.001),支持更准确的活检决策。
- 放射科医生或AI软件判为BI-RADS 2或3类的靶向超声病灶,本研究中病理均为良性。
局限性
- 本文为回顾性、横断面、单中心研究。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
INTRODUCTION: Targeted ultrasound is commonly used to identify lesions characterized on magnetic resonance imaging (MRI) that were not recognized on initial mammography or ultrasound and is especially valuable for guiding percutaneous biopsies. Although artificial intelligence (AI) algorithms have been used to differentiate benign from malignant breast lesions on ultrasound, their application in classifying lesions on targeted ultrasound has not yet been studied.
OBJECTIVE: To evaluate the performance of AI-based software in predicting malignancy risk in breast lesions identified on targeted ultrasound.
METHODS: This was a retrospective, cross-sectional, single-center study that included patients with breast lesions identified on MRI who underwent targeted ultrasound and percutaneous ultrasound-guided biopsy. The ultrasound findings were analyzed using AI-based software and subsequently correlated with the pathological results.
RESULTS: 334 lesions were evaluated, including 183 mass and 151 non-mass lesions. On histological analysis, there were 257 (76.9 %) benign lesions, and 77 (23.1 %) malignant. Both the AI software and radiologists demonstrated high sensitivity in predicting the malignancy risk of the lesions. The specificity was higher when evaluated by the radiologist using the AI software compared to the radiologist's evaluation alone (p < 0.001). All lesions classified as BI-RADS 2 or 3 on targeted ultrasound by the radiologist or the AI software (n = 72; 21.6 %) showed benign pathology results.
CONCLUSION: The AI software, when integrated into the radiologist's evaluation, demonstrated high diagnostic accuracy and improved specificity for both mass and non-mass lesions on targeted ultrasound, supporting more accurate biopsy decisions and potentially reducing false positives without missing cancers.
引用
- Lima IRM, Cruz RM, de Lima Rodrigues CL, et al. Performance of AI-Based software in predicting malignancy risk in breast lesions identified on targeted ultrasound. Eur J Radiol. 2025. PMID: 40743875. doi:10.1016/j.ejrad.2025.112339
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。