Division of Pulmonary, Allergy, and Critical Care Medicine, Department of Medicine, Stanford University, Stanford, CA, USA.
Imvaria, Inc, USA.
Respir Med. 2023 Nov-Dec;219:107428. doi: 10.1016/j.rmed.2023.107428. Epub 2023 Oct 13.
Non-invasive diagnosis of idiopathic pulmonary fibrosis (IPF) involves identification of usual interstitial pneumonia (UIP) pattern by computed tomography (CT) and exclusion of other known etiologies of interstitial lung disease (ILD). However, uncertainty in identification of radiologic UIP pattern leads to the continued need for invasive surgical biopsy. We thus developed and validated a machine learning algorithm using CT scans alone to augment non-invasive diagnosis of IPF.
The primary algorithm was a deep learning convolutional neural network (CNN) with model inputs of CT images only. The algorithm was trained to predict IPF among cases of ILD, with reference standard of multidisciplinary discussion (MDD) consensus diagnosis. The algorithm was trained using a multi-center dataset of more than 2000 cases of ILD. A US-based multi-site cohort (n = 295) was used for algorithm tuning, and external validation was performed with a separate dataset (n = 295) from European and South American sources.
In the tuning set, the model achieved an area under the receiver operating characteristic curve (AUC) of 0.87 (CI: 0.83-0.92) in differentiating IPF from other ILDs. Sensitivity and specificity were 0.67 (0.57-0.76) and 0.90 (0.83-0.95), respectively. By contrast, pre-recorded assessment prior to MDD diagnosis had sensitivity of 0.31 (0.23-0.42) and specificity of 0.92 (0.87-0.95). In the external test set, c-statistic was also 0.87 (0.83-0.91). Model performance was consistent across a variety of CT scanner manufacturers and slice thickness.
The presented deep learning algorithm demonstrated consistent performance in identifying IPF among cases of ILD using CT images alone and suggests generalization across CT manufacturers.
特发性肺纤维化（IPF）的无创诊断包括通过计算机断层扫描（CT）识别常见间质性肺炎（UIP）模式，并排除其他已知的间质性肺疾病（ILD）病因。然而，放射学 UIP 模式的识别存在不确定性，因此仍然需要进行有创的外科活检。因此，我们开发并验证了一种仅使用 CT 扫描的机器学习算法，以辅助 IPF 的无创诊断。
主要算法是一个深度学习卷积神经网络（CNN），模型输入仅为 CT 图像。该算法经过训练，可根据多学科讨论（MDD）共识诊断的参考标准，预测ILD 中的 IPF。该算法使用了一个包含 2000 多例ILD 病例的多中心数据集进行训练。一个基于美国的多站点队列（n=295）用于调整算法，另一个来自欧洲和南美洲的独立数据集（n=295）用于外部验证。
在调整集中，该模型在区分 IPF 与其他ILD 方面的受试者工作特征曲线下面积（AUC）为 0.87（CI：0.83-0.92）。敏感性和特异性分别为 0.67（0.57-0.76）和 0.90（0.83-0.95）。相比之下，在 MDD 诊断前预先记录的评估敏感性为 0.31（0.23-0.42），特异性为 0.92（0.87-0.95）。在外部测试集中，c 统计量也为 0.87（0.83-0.91）。该模型在各种 CT 扫描仪制造商和切片厚度方面的性能表现一致。
该深度学习算法在仅使用 CT 图像识别ILD 中的 IPF 方面表现出一致的性能，并表明在 CT 制造商之间具有推广性。
