Lasocha Bartlomiej, Pulyk Roman, Brzegowy Pawel, Latacz Pawel, Slowik Agnieszka, Popiela Tadeusz J
University Hospital New Seat (NSSU) Diagnostic Imaging Unit, University Hospital, 30-688 Krakow, Poland.
Chair of Neurology, Jagiellonian University Medical College, 31-008 Krakow, Poland.
J Clin Med. 2020 Oct 22;9(11):3383. doi: 10.3390/jcm9113383.
Our aim was to compare human and computer accuracy in reading medical images of acute stroke patients. We analyzed data of patients who underwent assessment of Alberta Stroke Program Early CT Score (ASPECTS) and CT Perfusion (CTP) via Rapid Processing of Perfusion and Diffusion (RAPID) software RAPID ASPECTS, and RAPID CTP), compared to radiologist reports and manual measurements. We compared volumes calculated by RAPID CTP software with those selected by scanner-equipped software (GE). For reference, follow-up images were manually assessed in accordance with the Alberta Stroke Program Early CT Score (ASPECTS) territories retrospectively. Although exact ASPECTS score agreement between the automatic and manual methods, and between each method and follow-up, was poor, crossing of the threshold for reperfusion therapy was characterized by an 80% match. CT perfusion analyses yielded only slight agreement (kappa = 0.193) in the qualification of patients for therapy. Either automatic or manual scoring methods of non-contrast images imply similar clinical decisions in real-world circumstances. However, volume measurements performed by fully automatic and manually assisted systems are not comparable. Thresholds devised and validated for computer algorithms are not compatible with measurements performed manually using other software and should not be applied to setups other than those with which they were developed.
我们的目的是比较人类和计算机在读取急性中风患者医学影像时的准确性。我们分析了通过灌注与扩散快速处理（RAPID）软件（RAPID ASPECTS和RAPID CTP）接受阿尔伯塔中风项目早期CT评分（ASPECTS）和CT灌注（CTP）评估的患者数据，并与放射科医生的报告和手动测量结果进行比较。我们将RAPID CTP软件计算的体积与配备扫描仪的软件（GE）选择的体积进行了比较。作为参考，随访图像根据阿尔伯塔中风项目早期CT评分（ASPECTS）区域进行了回顾性手动评估。尽管自动和手动方法之间以及每种方法与随访之间的ASPECTS评分确切一致性较差，但再灌注治疗阈值的跨越以80%的匹配度为特征。CT灌注分析在患者治疗资格的判定上仅产生了轻微的一致性（kappa = 0.193）。在实际临床环境中，非增强图像的自动或手动评分方法意味着相似的临床决策。然而，全自动和手动辅助系统进行的体积测量不可比。为计算机算法设计和验证的阈值与使用其他软件手动执行的测量不兼容，不应应用于其开发之外的设置。
