Troutt Hayden R, Huynh Kenneth N, Joshi Aditya, Ling Justin, Refugio Scott, Cramer Scott, Lopez Jasmine, Wei Katherine, Imanzadeh Amir, Chow Daniel S
UCI Health, Irvine, CA, USA.
J Imaging Inform Med. 2025 Jun 25. doi: 10.1007/s10278-025-01552-0.
The urgency to accelerate PE management and minimize patient risk has driven the development of artificial intelligence (AI) algorithms designed to provide a swift and accurate diagnosis in dedicated chest imaging (computed tomography pulmonary angiogram; CTPA) for suspected PE; however, the accuracy of AI algorithms in the detection of incidental PE in non-dedicated CT imaging studies remains unclear and untested. This study explores the potential for a commercial AI algorithm to identify incidental PE in non-dedicated contrast-enhanced CT chest imaging studies. The Viz PE algorithm was deployed to identify the presence of PE on 130 dedicated and 63 non-dedicated contrast-enhanced CT chest exams. The predictions for non-dedicated contrast-enhanced chest CT imaging studies were 90.48% accurate, with a sensitivity of 0.14 and specificity of 1.00. Our findings reflect that the Viz PE algorithm demonstrated an overall accuracy of 90.16%, with a specificity of 96% and a sensitivity of 41%. Although the high specificity is promising for ruling in PE, the low sensitivity highlights a limitation, as it indicates the algorithm may miss a substantial number of true-positive incidental PEs. This study demonstrates that commercial AI detection tools hold promise as integral support for detecting PE, particularly when there is a strong clinical indication for their use; however, current limitations in sensitivity, especially for incidental cases, underscore the need for ongoing radiologist oversight.
加速肺栓塞（PE）管理并将患者风险降至最低的紧迫性推动了人工智能（AI）算法的发展，这些算法旨在为疑似PE的专用胸部成像（计算机断层扫描肺动脉造影；CTPA）提供快速准确的诊断；然而，AI算法在非专用CT成像研究中检测偶发性PE的准确性仍不明确且未经测试。本研究探讨了一种商业AI算法在非专用对比增强胸部CT成像研究中识别偶发性PE的潜力。Viz PE算法被用于在130例专用和63例非专用对比增强胸部CT检查中识别PE的存在。非专用对比增强胸部CT成像研究的预测准确率为90.48%，敏感性为0.14，特异性为1.00。我们的研究结果表明，Viz PE算法的总体准确率为90.16%，特异性为96%，敏感性为41%。尽管高特异性有助于确诊PE，但低敏感性突出了一个局限性，因为这表明该算法可能会遗漏大量真正阳性的偶发性PE。本研究表明，商业AI检测工具有望成为检测PE的重要辅助手段，尤其是在有强烈临床使用指征时；然而，目前在敏感性方面的局限性，特别是对于偶发性病例，强调了持续进行放射科医生监督的必要性。
