Department of Radiology, Neuroradiology and Nuclear Medicine, Johannes Wesling University Hospital, Ruhr University Bochum, Bochum, Germany.
Sci Rep. 2023 Mar 5;13(1):3680. doi: 10.1038/s41598-023-30521-2.
The AI-Rad Companion Chest X-ray (AI-Rad, Siemens Healthineers) is an artificial-intelligence based application for the analysis of chest X-rays. The purpose of the present study is to evaluate the performance of the AI-Rad. In total, 499 radiographs were retrospectively included. Radiographs were independently evaluated by radiologists and the AI-Rad. Findings indicated by the AI-Rad and findings described in the written report (WR) were compared to the findings of a ground truth reading (consensus decision of two radiologists after assessing additional radiographs and CT scans). The AI-Rad can offer superior sensitivity for the detection of lung lesions (0.83 versus 0.52), consolidations (0.88 versus 0.78) and atelectasis (0.54 versus 0.43) compared to the WR. However, the superior sensitivity is accompanied by higher false-detection-rates. The sensitivity of the AI-Rad for the detection of pleural effusions is lower compared to the WR (0.74 versus 0.88). The negative-predictive-values (NPV) of the AI-Rad for the detection of all pre-defined findings are on a high level and comparable to the WR. The seemingly advantageous high sensitivity of the AI-Rad is partially offset by the disadvantage of a high false-detection-rate. At the current stage of development, therefore, the high NPVs may be the greatest benefit of the AI-Rad giving radiologists the possibility to re-insure their own negative search for pathologies and thus boosting their confidence in their reports.
人工智能辅助胸部 X 射线（AI-Rad，西门子医疗）是一种基于人工智能的胸部 X 射线分析应用程序。本研究旨在评估 AI-Rad 的性能。共回顾性纳入 499 张射线照片。射线照片由放射科医生和 AI-Rad 独立评估。AI-Rad 指示的发现和书面报告（WR）中描述的发现与地面实况阅读（在评估其他射线照片和 CT 扫描后，由两名放射科医生做出共识决定）的发现进行比较。与 WR 相比，AI-Rad 可以更灵敏地检测到肺部病变（0.83 对 0.52）、实变（0.88 对 0.78）和肺不张（0.54 对 0.43）。然而，更高的灵敏度伴随着更高的假阳性率。与 WR 相比，AI-Rad 检测胸腔积液的灵敏度较低（0.74 对 0.88）。AI-Rad 对所有预定义发现的阴性预测值（NPV）均处于较高水平，与 WR 相当。因此，AI-Rad 看似有利的高灵敏度部分被高假阳性率所抵消。在目前的发展阶段，高 NPV 可能是 AI-Rad 的最大优势，它使放射科医生有可能重新确认他们自己对病变的阴性搜索，从而增强他们对报告的信心。
