AI驱动的前列腺MRI分诊流程模拟评估:临床显著性前列腺癌诊断
本研究基于国际AI验证研究数据,回顾性模拟了AI驱动的前列腺MRI分诊流程,对500例疑似csPCa患者中的400例进行测试,AI系统以高置信度阈值评估部分检查,其余转交放射科医生。结论认为该AI分诊系统可提高模拟诊断流程效率而不影响诊断准确性。
要点
- AI分诊流程在保持敏感度相当的同时,将特异度从57.7%提升至69.2%。
- AI独立分诊并诊断了49%的检查,其敏感度和特异度均达94.7%。
- AI分诊可减少放射科医生阅片量,潜在降低工作负荷。
英文原文(摘要)· 点击展开
Purpose To simulate an artificial intelligence (AI)-driven triaging workflow in which an AI system, using high-confidence thresholds, assesses a subset of prostate MRI examinations for clinically significant prostate cancer (csPCa), compare the assessment with stand-alone radiologists, and evaluate the number of examinations triaged by the AI to estimate potential workload reduction. Materials and Methods Data from an international AI confirmatory study (February 2022-November 2023) were used in this retrospective study. MRI examinations of 500 men with suspected csPCa from four European centers were included. Exclusion criteria were prior prostate treatment, prior csPCa, or considerable imaging artifacts. AI-triaging thresholds were calibrated on 100 examinations. The AI system assessed examinations exceeding high-specificity or high-sensitivity thresholds, with the remaining examinations deferred to radiologists. The workflow was simulated on 400 examinations, including examinations from an external site, incorporating assessments from 62 radiologists. Reference standards were histopathology and/or 3 or more years of follow-up. Sensitivity and specificity of the triaging workflow were compared with the conventional workflow using multireader, multicase analysis of variance. Results Among the 400 patients (median age, 66 years; IQR, 60-69 years) included for testing, radiologists achieved a sensitivity of 89.4% (95% CI: 85.8, 93.1) and specificity of 57.7% (95% CI: 52.3, 63.1). The AI-driven pathway maintained comparable sensitivity (89.0%; 95% CI: 85.0, 93.0; P = .36) but improved specificity by 11.5%, reaching 69.2% (95% CI: 64.4, 74.0; P < .001). The AI system triaged and diagnosed 195 of 400 (49%; 95% CI: 173, 216) examinations with sensitivity of 94.7% (95% CI: 89.5, 99.9) and specificity of 94.7% (95% CI: 90.5, 98.9). Conclusion Triaging by this AI system improved simulated diagnostic workflow efficiency without compromising diagnostic accuracy for csPCa. Keywords: Prostate, MRI, Localization, Oncology, Comparative Studies, Diagnosis Supplemental material is available for this article. ClinicalTrials.gov registration no. NCT05489341 © RSNA, 2026.
引用
- Twilt JJ, Saha A, Bosma JS, et al. Evaluating an AI-driven Triaging Workflow for MRI-based Clinically Significant Prostate Cancer Diagnosis: A Simulation Study. Radiol Imaging Cancer. 2026. PMID: 41960994. doi:10.1148/rycan.250461
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。