基于双参数MRI的人工智能对临床显著性前列腺癌的诊断效能:系统综述与Meta分析
该系统综述与Meta分析纳入19项研究,评估基于双参数MRI的人工智能算法诊断临床显著性前列腺癌的效能。内部验证集4594例,平均敏感度0.88、特异度0.79、AUC 0.91;外部验证集795例,平均敏感度0.85、特异度0.83、AUC 0.91;放射科医师队列897例,平均AUC 0.78。作者认为AI诊断效能优异,可能优于传统影像解读,但研究间异质性显著且放射科医师与外部验证集样本量有限,需谨慎对待,并需进一步前瞻性研究。
要点
- 基于双参数MRI的AI诊断临床显著性前列腺癌内部验证平均AUC 0.91(95%CI 0.88-0.93)。
- 外部验证平均AUC 0.91(95%CI 0.88-0.93),与内部验证一致。
- 放射科医师评估平均AUC 0.78(95%CI 0.74-0.81),低于AI模型。
核心结论
- 基于双参数MRI的AI对临床显著性前列腺癌表现出优异的诊断效能,是一种有前景的无创方法,可能优于传统影像解读;但研究间异质性显著且放射科医师与外部验证集样本量有限,需谨慎对待,需进一步前瞻性研究以证实这些发现并探究临床适用性。
研究设计与样本
- 遵循PRISMA-DTA指南的系统综述与Meta分析,检索PubMed、Embase和Web of Science数据库,纳入基于双参数MRI的AI算法诊断临床显著性前列腺癌的研究,采用双变量随机效应模型合并敏感度、特异度和AUC,用I²统计量评估异质性并按AI方法学、AI模型、研究设计和地理区域进行亚组分析。
诊断标准与关键数值
- 内部验证集平均敏感度0.88(95%CI 0.84-0.92)、平均特异度0.79(95%CI 0.73-0.84)、平均AUC 0.91(95%CI 0.88-0.93);外部验证集平均敏感度0.85(95%CI 0.78-0.90)、平均特异度0.83(95%CI 0.69-0.91)、平均AUC 0.91(95%CI 0.88-0.93);放射科医师评估平均AUC 0.78(95%CI 0.74-0.81)。
临床应用与阅片要点
- 作者认为基于双参数MRI的AI诊断临床显著性前列腺癌效能优异,可能优于传统影像解读,但研究间异质性显著且放射科医师与外部验证集样本量有限,需谨慎对待,需进一步前瞻性研究以探究临床适用性。
局限性
- 研究间存在显著异质性。
- 放射科医师队列和外部验证集的样本量有限。
- 需进一步前瞻性研究以证实这些发现并探究临床适用性。
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
OBJECTIVES: This meta-analysis aimed to systematically evaluate the diagnostic performance of artificial intelligence (AI) applied to biparametric magnetic resonance imaging (bpMRI) for identifying clinically significant prostate cancer (csPCa).
METHODS: A comprehensive systematic review was conducted following PRISMA-DTA guidelines, searching PubMed, Embase, and Web of Science databases. Studies focus on AI algorithms based on bpMRI in diagnosis csPCa were included. Bivariate random-effects models synthesized sensitivity, specificity, and area under the curve (AUC). Heterogeneity was assessed using I² statistics, with subgroup analyses exploring variations across AI methodologies, AI models, study designs, and geographical regions.
RESULTS: Nineteen studies were included, encompassing 4594 patients in internal validation sets, 795 in external validation sets, and 897 in radiologist cohorts. AI models based on bpMRI exhibited notable diagnostic performance, with internal validation revealing an average sensitivity of 0.88 (95% CI: 0.84-0.92), average specificity of 0.79 (95% CI: 0.73-0.84), and an average AUC of 0.91 (95% CI: 0.88-0.93). External validation confirmed these results with a average sensitivity of 0.85 (95% CI: 0.78-0.90), average specificity of 0.83 (95% CI: 0.69-0.91), and an average AUC of 0.91 (95% CI: 0.88-0.93). In contrast, radiologist assessments showed lower performance with an average AUC of 0.78 (95% CI: 0.74-0.81).
CONCLUSION: AI applied to bpMRI demonstrates excellent diagnostic performance for csPCa, representing a promising noninvasive approach that may potentially outperform traditional radiological interpretations. However, notable heterogeneity across studies and limited sample size for radiologists and external validation sets suggests the need for caution. To substantiate these findings and investigate clinical applicability, additional prospective studies are essential.
引用
- Yan G, Wang Y, Chen L, et al. Diagnostic Performance of Artificial Intelligence Based on Biparametric MRI for Clinically Significant Prostate Cancer: A Systematic Review and Meta-analysis. Acad Radiol. 2025. PMID: 40058989. doi:10.1016/j.acra.2025.02.044
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。