Elhakim Mohammad T, Stougaard Sarah W, Graumann Ole, Nielsen Mads, Gerke Oke, Larsen Lisbet B, Rasmussen Benjamin S B
From the Department of Radiology (M.T.E., L.B.L., B.S.B.R.), Department of Nuclear Medicine (O. Gerke), and CAI-X-Centre for Clinical Artificial Intelligence (B.S.B.R.), Odense University Hospital, Kløvervænget 10, Entrance 112, 2nd Floor, 5000 Odense C, Denmark; Department of Clinical Research, Research and Innovation Unit of Radiology, University of Southern Denmark, Odense, Denmark (M.T.E., S.W.S., O. Graumann, O. Gerke, B.S.B.R.); Department of Radiology, Aarhus University Hospital, Aarhus, Denmark (O. Graumann); Department of Clinical Research, Aarhus University, Aarhus, Denmark (O. Graumann); and Department of Computer Science, University of Copenhagen, Copenhagen, Denmark (M.N.).
Radiol Artif Intell. 2024 Nov;6(6):e230529. doi: 10.1148/ryai.230529.
Mammography screening supported by deep learning-based artificial intelligence (AI) solutions can potentially reduce workload without compromising breast cancer detection accuracy, but the site of deployment in the workflow might be crucial. This retrospective study compared three simulated AI-integrated screening scenarios with standard double reading with arbitration in a sample of 249 402 mammograms from a representative screening population. A commercial AI system replaced the first reader (scenario 1: integrated AI), the second reader (scenario 2: integrated AI), or both readers for triaging of low- and high-risk cases (scenario 3: integrated AI). AI threshold values were chosen based partly on previous validation and setting the screen-read volume reduction at approximately 50% across scenarios. Detection accuracy measures were calculated. Compared with standard double reading, integrated AI showed no evidence of a difference in accuracy metrics except for a higher arbitration rate (+0.99%, < .001). Integrated AI had lower sensitivity (-1.58%, < .001), negative predictive value (NPV) (-0.01%, < .001), and recall rate (-0.06%, = .04) but a higher positive predictive value (PPV) (+0.03%, < .001) and arbitration rate (+1.22%, < .001). Integrated AI achieved higher sensitivity (+1.33%, < .001), PPV (+0.36%, = .03), and NPV (+0.01%, < .001) but lower arbitration rate (-0.88%, < .001). Replacing one or both readers with AI seems feasible; however, the site of application in the workflow can have clinically relevant effects on accuracy and workload. Mammography, Breast, Neoplasms-Primary, Screening, Epidemiology, Diagnosis, Convolutional Neural Network (CNN) Published under a CC BY 4.0 license.
基于深度学习的人工智能（AI）解决方案支持的乳腺钼靶筛查有可能在不影响乳腺癌检测准确性的情况下减少工作量，但在工作流程中的部署位置可能至关重要。这项回顾性研究在来自代表性筛查人群的249402例乳腺钼靶检查样本中，将三种模拟的人工智能集成筛查方案与标准双重读片并进行仲裁的方案进行了比较。一个商业人工智能系统取代了第一个读片者（方案1：集成人工智能）、第二个读片者（方案2：集成人工智能），或同时取代两个读片者对低风险和高风险病例进行分流（方案3：集成人工智能）。人工智能阈值部分基于先前的验证来选择，并将各方案的筛查读片量减少设定为约50%。计算了检测准确性指标。与标准双重读片相比，除了仲裁率更高（+0.99%，P<0.001）外，集成人工智能在准确性指标上没有差异的证据。集成人工智能的灵敏度较低（-1.58%，P<0.001）、阴性预测值（NPV）较低（-0.01%，P<0.001）和召回率较低（-0.06%，P=0.04），但阳性预测值（PPV）较高（+0.03%，P<0.001）和仲裁率较高（+1.22%，P<0.001）。集成人工智能实现了更高的灵敏度（+1.33%，P<0.001）、PPV（+0.36%，P=0.03）和NPV（+0.01%，P<0.001），但仲裁率较低（-0.88%，P<0.001）。用人工智能取代一个或两个读片者似乎是可行的；然而，在工作流程中的应用位置可能会对准确性和工作量产生临床相关影响。乳腺钼靶、乳腺、原发性肿瘤、筛查、流行病学、诊断、卷积神经网络（CNN） 依据知识共享署名4.0许可协议发布。
