弥散加权成像在检出人工智能乳腺X线摄影漏诊乳腺癌中的附加价值
本回顾性研究纳入414例病理确诊乳腺癌女性,评估AI-CAD漏诊的127例(30.7%)癌症特征,发现致密乳腺和肿瘤≤2 cm是漏诊独立预测因素;单独DWI对AI漏诊癌的检出率两位阅片者分别为83.5%和79.5%,一致性κ=0.61,支持DWI作为AI筛查的辅助分诊工具,但回顾性、仅癌症设计限制推广性。
要点
- AI-CAD漏诊了30.7%(127/414)的乳腺癌,致密乳腺和肿瘤≤2 cm是独立预测因素。
- 单独DWI可检出83.5%和79.5%的AI漏诊癌,两位阅片者一致性κ=0.61。
- DWI对乳腺X线隐匿或>1 cm肿瘤检出有效,但对亚厘米病变敏感度下降。
核心结论
- 单独DWI可检出大多数AI-CAD漏诊的乳腺癌,支持其作为AI筛查中辅助分诊的潜在作用,尤其适用于致密乳腺和乳腺X线隐匿病变。
- 回顾性、仅癌症设计限制推广性,强调需前瞻性多中心筛查试验验证。
研究设计与样本
- 回顾性研究纳入414例病理确诊乳腺癌女性(平均年龄55.3岁),均行术前乳腺X线摄影、MRI含DWI及手术。
- 以AI-CAD异常评分>10且正确局部化为AI检出,否则为AI漏诊;比较两组临床病理及影像特征。
- 两位放射科医师独立阅DWI并对AI漏诊癌赋6点Likert恶性置信度评分(≥3为阳性),分析阅片者间一致性及诊断性能。
关键影像表现/征象
- AI-CAD漏诊127/414例乳腺癌(30.7%)。
- 多变量回归显示致密乳腺(校正OR=1.619;p=0.049)和肿瘤≤2 cm(校正OR=4.698;p<0.001)是AI漏诊癌的独立预测因素。
诊断标准与关键数值
- AI-CAD异常评分>10且正确局部化定义为AI检出,否则为AI漏诊。
- DWI阅片采用6点Likert恶性置信度评分,≥3分视为阳性。
- 单独DWI对AI漏诊癌的检出率:阅片者1为83.5%,阅片者2为79.5%,阅片者间一致性κ=0.61。
临床应用与阅片要点
- DWI对乳腺X线隐匿或>1 cm的AI漏诊肿瘤检出有效,但亚厘米病变敏感度下降。
- 作者认为单独DWI可检出大多数AI-CAD漏诊乳腺癌,支持其作为AI筛查中辅助分诊的潜在作用,尤其适用于致密乳腺和乳腺X线隐匿病变。
局限性
- 回顾性、仅癌症设计限制推广性。
- 需前瞻性多中心筛查试验验证。
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
OBJECTIVE: To evaluate breast cancers missed by artificial intelligence-based computer-aided diagnosis (AI-CAD) in women newly diagnosed with breast cancer, identify factors associated with these missed cases, and assess the potential diagnostic value of standalone diffusion-weighted imaging (DWI) in detecting cancers overlooked by AI-CAD.
MATERIALS AND METHODS: This retrospective study included 414 women (mean age, 55.3 years) with pathologically confirmed breast cancer who underwent preoperative mammography, MRI with DWI, and surgery. Cancers were classified as AI-detected if the lesion had an abnormality score greater than 10 and was correctly localized by AI-CAD; otherwise, they were categorized as AI-missed. Clinicopathologic and imaging features were compared between groups. Two radiologists independently reviewed DWI of AI-missed cancers and assigned malignancy confidence scores using a 6-point Likert-type scale (≥3 considered positive). Interobserver agreement and diagnostic performance were analyzed.
RESULTS: AI-CAD missed 127 of 414 breast cancers (30.7%). Multivariate regression analysis identified dense breasts (adjusted OR = 1.619; p = 0.049) and tumor size ≤ 2 cm (adjusted OR = 4.698; p < 0.001) as independent predictors of AI-missed cancer. Standalone DWI detected 83.5% and 79.5% of AI-missed cancers for Radiologists 1 and 2, respectively, with substantial agreement (κ = 0.61). DWI was effective in detecting mammographically occult or >1 cm tumors, but sensitivity declined for subcentimeter lesions.
CONCLUSION: Standalone DWI detects the majority of breast cancers missed by AI-CAD, supporting its potential role as a triage adjunct in AI-based screening, particularly for dense breasts and mammographically occult lesions. However, the retrospective, cancer-only design limits generalizability, highlighting the need for prospective multicenter screening trials for validation.
引用
- Kim JY, Kim JJ, Lee HJ, et al. Added value of diffusion-weighted imaging in detecting breast cancer missed by artificial intelligence-based mammography. Radiol Med. 2025. PMID: 41324774. doi:10.1007/s11547-025-02161-1
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。