From the Department of Neuroradiology (M.A.M., C.J.P., H.M., M.B., G.B., P.V.), Department of Neuroradiology, Division for Computational Neuroimaging (M.A.M., C.J.P., H.M., G.B., P.V.), and Department of Neurology (W.W.), Heidelberg University Hospital, Im Neuenheimer Feld 400, 69120 Heidelberg, Germany; Department of Neurosurgery, University Hospital Munich LMU, Munich, Germany (J.C.T.); and Department of Neurology, Clinical Neuroscience Center, University Hospital Zurich and University of Zurich, Zurich, Switzerland (M.W.).
Radiol Artif Intell. 2024 Jan;6(1):e230095. doi: 10.1148/ryai.230095.
Purpose To develop a fully automated device- and sequence-independent convolutional neural network (CNN) for reliable and high-throughput labeling of heterogeneous, unstructured MRI data. Materials and Methods Retrospective, multicentric brain MRI data (2179 patients with glioblastoma, 8544 examinations, 63 327 sequences) from 249 hospitals and 29 scanner types were used to develop a network based on ResNet-18 architecture to differentiate nine MRI sequence types, including T1-weighted, postcontrast T1-weighted, T2-weighted, fluid-attenuated inversion recovery, susceptibility-weighted, apparent diffusion coefficient, diffusion-weighted (low and high value), and gradient-recalled echo T2*-weighted and dynamic susceptibility contrast-related images. The two-dimensional-midsection images from each sequence were allocated to training or validation (approximately 80%) and testing (approximately 20%) using a stratified split to ensure balanced groups across institutions, patients, and MRI sequence types. The prediction accuracy was quantified for each sequence type, and subgroup comparison of model performance was performed using χ tests. Results On the test set, the overall accuracy of the CNN (ResNet-18) ensemble model among all sequence types was 97.9% (95% CI: 97.6, 98.1), ranging from 84.2% for susceptibility-weighted images (95% CI: 81.8, 86.6) to 99.8% for T2-weighted images (95% CI: 99.7, 99.9). The ResNet-18 model achieved significantly better accuracy compared with ResNet-50 despite its simpler architecture (97.9% vs 97.1%; ≤ .001). The accuracy of the ResNet-18 model was not affected by the presence versus absence of tumor on the two-dimensional-midsection images for any sequence type ( > .05). Conclusion The developed CNN () reliably differentiates nine types of MRI sequences within multicenter and large-scale population neuroimaging data and may enhance the speed, accuracy, and efficiency of clinical and research neuroradiologic workflows. MR-Imaging, Neural Networks, CNS, Brain/Brain Stem, Computer Applications-General (Informatics), Convolutional Neural Network (CNN), Deep Learning Algorithms, Machine Learning Algorithms © RSNA, 2023.
开发一种完全自动化的、与设备和序列无关的卷积神经网络（CNN），用于可靠、高通量地对异质、非结构化的 MRI 数据进行标注。
本研究回顾性分析了来自 249 家医院和 29 种扫描仪类型的 2179 例脑胶质母细胞瘤患者的 8544 次检查、63327 个序列的脑 MRI 数据，用于开发基于 ResNet-18 架构的网络，以区分 9 种 MRI 序列类型，包括 T1 加权、对比后 T1 加权、T2 加权、液体衰减反转恢复、磁化率加权、表观扩散系数、扩散加权（低值和高值）、梯度回波 T2*-加权和动态磁敏感对比相关图像。使用分层分割将每个序列的二维中截面图像分配到训练或验证（约 80%）和测试（约 20%）中，以确保机构、患者和 MRI 序列类型之间的分组平衡。对每个序列类型的预测准确性进行量化，并使用 χ 检验进行模型性能的亚组比较。
在测试集中，CNN（ResNet-18）集成模型在所有序列类型中的总体准确率为 97.9%（95%CI：97.6，98.1），范围从磁化率加权图像的 84.2%（95%CI：81.8，86.6）到 T2 加权图像的 99.8%（95%CI：99.7，99.9）。尽管 ResNet-50 模型的结构更简单，但 ResNet-18 模型的准确率显著更高（97.9%比 97.1%； ≤.001）。对于任何序列类型，二维中截面图像上是否存在肿瘤对 ResNet-18 模型的准确率均无影响（ >.05）。
该研究开发的 CNN 可在多中心、大规模人群神经影像学数据中可靠地区分 9 种 MRI 序列类型，可能会提高临床和研究神经放射学工作流程的速度、准确性和效率。
磁共振成像、神经网络、中枢神经系统、脑/脑干、计算机应用一般（信息学）、卷积神经网络（CNN）、深度学习算法、机器学习算法
©RSNA，2023。
