Abstract

Computed tomography (CT) is one of the most efficient diagnostic methods for rapid diagnosis of the widespread COVID-19. However, reading CT films brings a lot of concentration and time for doctors. Therefore, it is necessary to develop an automatic CT image diagnosis system to assist doctors in diagnosis. Previous studies devoted to COVID-19 in the past months focused mostly on discriminating COVID-19 infected patients from healthy persons and/or bacterial pneumonia patients, and have ignored typical viral pneumonia since it is hard to collect samples for viral pneumonia that is less frequent in adults. In addition, it is much more challenging to discriminate COVID-19 from typical viral pneumonia as COVID-19 is also a kind of virus. In this study, we have collected CT images of 262, 100, 219, and 78 persons for COVID-19, bacterial pneumonia, typical viral pneumonia, and healthy controls, respectively. To the best of our knowledge, this was the first study of quaternary classification to include also typical viral pneumonia. To effectively capture the subtle differences in CT images, we have constructed a new model by combining the ResNet50 backbone with SE blocks that was recently developed for fine image analysis. Our model was shown to outperform commonly used baseline models, achieving an overall accuracy of 0.94 with AUC of 0.96, recall of 0.94, precision of 0.95, and F1-score of 0.94. The model is available in https://github.com/Zhengfudan/COVID-19-Diagnosis-and-Pneumonia-Classification.

Graphic Abstract

中文翻译：

---

通过深度学习根据 CT 图像准确区分 COVID-19 与病毒性和细菌性肺炎

摘要

计算机断层扫描 (CT) 是快速诊断广泛传播的 COVID-19 的最有效诊断方法之一。然而，阅读CT片给医生带来了大量的注意力和时间。因此，有必要开发一种自动CT图像诊断系统，以辅助医生进行诊断。在过去几个月里，以前针对 COVID-19 的研究主要集中在区分 COVID-19 感染患者与健康人和/或细菌性肺炎患者，而忽略了典型的病毒性肺炎，因为很难收集不太常见的病毒性肺炎样本在成年人中。此外，由于 COVID-19 也是一种病毒，因此将 COVID-19 与典型的病毒性肺炎区分开来更具挑战性。在这项研究中，我们收集了 262、100、219 和 78 人的 CT 图像，用于 COVID-19、细菌性肺炎、分别为典型的病毒性肺炎和健康对照。据我们所知，这是第一个还包括典型病毒性肺炎的第四纪分类研究。为了有效地捕获 CT 图像中的细微差异，我们通过将 ResNet50 主干与最近开发的用于精细图像分析的 SE 块相结合，构建了一个新模型。我们的模型被证明优于常用的基线模型，总体准确度为 0.94，AUC 为 0.96，召回率为 0.94，精度为 0.95，F1 分数为 0.94。该模型可在 https://github.com/Zhengfudan/COVID-19-Diagnosis-and-Pneumonia-Classification 中找到。为了有效地捕获 CT 图像中的细微差异，我们通过将 ResNet50 主干与最近开发的用于精细图像分析的 SE 块相结合，构建了一个新模型。我们的模型被证明优于常用的基线模型，总体准确度为 0.94，AUC 为 0.96，召回率为 0.94，精度为 0.95，F1 分数为 0.94。该模型可在 https://github.com/Zhengfudan/COVID-19-Diagnosis-and-Pneumonia-Classification 中找到。为了有效地捕获 CT 图像中的细微差异，我们通过将 ResNet50 主干与最近开发的用于精细图像分析的 SE 块相结合，构建了一个新模型。我们的模型被证明优于常用的基线模型，总体准确度为 0.94，AUC 为 0.96，召回率为 0.94，精度为 0.95，F1 分数为 0.94。该模型可在 https://github.com/Zhengfudan/COVID-19-Diagnosis-and-Pneumonia-Classification 中找到。

图形摘要