Background: Prostate cancer is one of the most common malignancies in men and a leading cause of cancer-related mortality. Early detection is essential to ensure curative treatment and favorable outcomes, but traditional diagnostic approaches—such as serum prostate-specific antigen (PSA) testing, digital rectal examination (DRE), and histopathological confirmation following biopsy—are limited by suboptimal accuracy and variability. Multiparametric magnetic resonance imaging (mpMRI) has improved diagnostic performance but remains highly dependent on reader expertise. Artificial intelligence (AI) offers promising opportunities to enhance diagnostic accuracy, reproducibility, and efficiency in prostate cancer detection. Objective: To evaluate the diagnostic accuracy and reporting timeliness of AI-based technologies compared with conventional diagnostic methods in the early detection of prostate cancer. Methods: Following PRISMA 2020 guidelines, PubMed, Scopus, Web of Science, and Cochrane Library were searched for studies published between January 2015 and April 2025. Eligible designs included randomized controlled trials, cohort, case–control, and pilot studies applying AI-based technologies to early prostate cancer diagnosis. Data on AUC-ROC, sensitivity, specificity, predictive values, diagnostic odds ratio (DOR), and time-to-reporting were narratively synthesized due to heterogeneity. Risk of bias was assessed using the QUADAS-AI tool. Results: Twenty-three studies involving 23,270 patients were included. AI-based technologies achieved a median AUC-ROC of 0.88 (range 0.70–0.93), with median sensitivity and specificity of 0.86 and 0.83, respectively. Compared with radiologists, AI or AI-assisted readings improved or matched diagnostic accuracy, reduced inter-reader variability, and decreased reporting time by up to 56%. Conclusions: AI-based technologies show strong diagnostic performance in early prostate cancer detection. However, methodological heterogeneity and limited standardization restrict generalizability. Large-scale prospective trials are required to validate clinical integration.
背景：前列腺癌是男性最常见的恶性肿瘤之一，也是癌症相关死亡的主要原因。早期检测对于确保治愈性治疗和良好预后至关重要，但传统的诊断方法——如血清前列腺特异性抗原（PSA）检测、直肠指检（DRE）以及活检后的组织病理学确认——受限于其准确性和一致性欠佳。多参数磁共振成像（mpMRI）提高了诊断性能，但仍高度依赖阅片者的专业水平。人工智能（AI）为提高前列腺癌检测的诊断准确性、可重复性和效率提供了有前景的机遇。目的：评估基于AI的技术与传统诊断方法在前列腺癌早期检测中的诊断准确性和报告及时性。方法：遵循PRISMA 2020指南，检索了PubMed、Scopus、Web of Science和Cochrane Library数据库中2015年1月至2025年4月发表的研究。符合条件的研究设计包括将基于AI的技术应用于早期前列腺癌诊断的随机对照试验、队列研究、病例对照研究和初步研究。由于存在异质性，对AUC-ROC、敏感性、特异性、预测值、诊断比值比（DOR）和报告时间的数据进行了叙述性综合。使用QUADAS-AI工具评估偏倚风险。结果：共纳入23项研究，涉及23,270名患者。基于AI的技术获得的中位AUC-ROC为0.88（范围0.70–0.93），中位敏感性和特异性分别为0.86和0.83。与放射科医生相比，AI或AI辅助阅片提高或匹配了诊断准确性，减少了阅片者间的差异，并将报告时间缩短了高达56%。结论：基于AI的技术在前列腺癌早期检测中显示出强大的诊断性能。然而，方法学的异质性和有限的标准化限制了其普适性。需要进行大规模前瞻性试验以验证其临床整合。
肿瘤(癌症)患者之家