From the Departments of Electrical Computer Engineering (Y.W., S.L., A.A., A.K.K.) and Computer Science (A.K.K.), Northwestern University, Evanston, Ill; Departments of Radiology (M.I., S.B., D.R.C., N.S., M.D., T.A.H., E.J.R., T.B.P., A.K.K., V.B.H.) and Neurology (A.M.N.), Northwestern University Feinberg School of Medicine, 676 N St. Clair St, Ste 1400, Chicago, IL 60611; Shirley Ryan AbilityLab, Chicago, Ill (S.B.); Department of Radiology, Indiana University Health, Indianapolis, Ind (J.Z.); Department of Radiology, Medical College of Wisconsin, Milwaukee, Wis (E.J.T.); Department of Medical Imaging, McMaster University, Hamilton, Ontario, Canada, (S.T.H.); and Department of Radiology, Mount Sinai Medical Center, Miami Beach, Fla (K.M.P.).
Radiol Artif Intell. 2024 Nov;6(6):e230296. doi: 10.1148/ryai.230296.
Purpose To develop a highly generalizable weakly supervised model to automatically detect and localize image-level intracranial hemorrhage (ICH) by using study-level labels. Materials and Methods In this retrospective study, the proposed model was pretrained on the image-level Radiological Society of North America dataset and fine-tuned on a local dataset by using attention-based bidirectional long short-term memory networks. This local training dataset included 10 699 noncontrast head CT scans in 7469 patients, with ICH study-level labels extracted from radiology reports. Model performance was compared with that of two senior neuroradiologists on 100 random test scans using the McNemar test, and its generalizability was evaluated on an external independent dataset. Results The model achieved a positive predictive value (PPV) of 85.7% (95% CI: 84.0, 87.4) and an area under the receiver operating characteristic curve of 0.96 (95% CI: 0.96, 0.97) on the held-out local test set ( = 7243, 3721 female) and 89.3% (95% CI: 87.8, 90.7) and 0.96 (95% CI: 0.96, 0.97), respectively, on the external test set ( = 491, 178 female). For 100 randomly selected samples, the model achieved performance on par with two neuroradiologists, but with a significantly faster ( < .05) diagnostic time of 5.04 seconds per scan (vs 86 seconds and 22.2 seconds for the two neuroradiologists, respectively). The model's attention weights and heatmaps visually aligned with neuroradiologists' interpretations. Conclusion The proposed model demonstrated high generalizability and high PPVs, offering a valuable tool for expedited ICH detection and prioritization while reducing false-positive interruptions in radiologists' workflows. Computer-Aided Diagnosis (CAD), Brain/Brain Stem, Hemorrhage, Convolutional Neural Network (CNN), Transfer Learning © RSNA, 2024 See also the commentary by Akinci D'Antonoli and Rudie in this issue.
目的 通过使用研究级标签，开发一种高度可推广的弱监督模型，自动检测和定位图像级颅内出血（ICH）。
材料和方法 在这项回顾性研究中，所提出的模型在放射学会北美数据集上进行了预训练，并通过基于注意力的双向长短时记忆网络在本地数据集上进行了微调。这个本地训练数据集包括了 7469 名患者的 10699 次非对比头部 CT 扫描，ICH 的研究级标签是从放射学报告中提取的。通过麦克内马尔检验，将该模型在 100 次随机测试扫描中的性能与两名资深神经放射学家进行了比较，并在一个独立的外部数据集上评估了其泛化能力。
结果 在保留的本地测试集中（n=7243，3721 名女性），该模型对 100 个随机测试样本的阳性预测值（PPV）为 85.7%（95%CI：84.0，87.4），接收器操作特征曲线下面积为 0.96（95%CI：0.96，0.97）；在外部测试集（n=491，178 名女性）中，PPV 分别为 89.3%（95%CI：87.8，90.7）和 0.96（95%CI：0.96，0.97）。对于 100 个随机选择的样本，该模型与两名神经放射学家的表现相当，但诊断时间明显更快（<0.05），每个扫描为 5.04 秒（而两名神经放射学家的诊断时间分别为 86 秒和 22.2 秒）。该模型的注意力权重和热图与神经放射学家的解释视觉上一致。
结论 该模型表现出高度的可推广性和高的 PPV，为加快 ICH 的检测和优先级排序提供了一个有价值的工具，同时减少了放射科工作流程中的假阳性中断。计算机辅助诊断（CAD），脑/脑干，出血，卷积神经网络（CNN），迁移学习
RSNA，2024 年 也可参见本期 Akinci D'Antonoli 和 Rudie 的评论。
