Department of Medical Imaging, Guangdong Second Provincial General Hospital, Guangzhou, 510317, People's Republic of China.
Radiology Department, Xiao Chang First People's Hospital, Hubei, People's Republic of China.
Sci Rep. 2021 Sep 9;11(1):17885. doi: 10.1038/s41598-021-97497-9.
We propose a classification method using the radiomics features of CT chest images to identify patients with coronavirus disease 2019 (COVID-19) and other pneumonias. The chest CT images of two groups of participants (90 COVID-19 patients who were confirmed as positive by nucleic acid test of RT-PCR and 90 other pneumonias patients) were collected, and the two groups of data were manually drawn to outline the region of interest (ROI) of pneumonias. The radiomics method was used to extract textural features and histogram features of the ROI and obtain a radiomics features vector from each sample. Then, we divided the data into two independent radiomic cohorts for training (70 COVID-19 patients and 70 other pneumonias patients), and validation (20 COVID-19 patients and 20 other pneumonias patients) by using support vector machine (SVM). This model used 20 rounds of tenfold cross-validation for training. Finally, single-shot testing of the final model was performed on the independent validation cohort. In the COVID-19 patients, correlation analysis (multiple comparison correction-Bonferroni correction, P < 0.05/7) was also conducted to determine whether the textural and histogram features were correlated with the laboratory test index of blood, i.e., blood oxygen, white blood cell, lymphocytes, neutrophils, C-reactive protein, hypersensitive C-reactive protein, and erythrocyte sedimentation rate. The final model showed good discrimination on the independent validation cohort, with an accuracy of 89.83%, sensitivity of 94.22%, specificity of 85.44%, and AUC of 0.940. This proved that the radiomics features were highly distinguishable, and this SVM model can effectively identify and diagnose patients with COVID-19 and other pneumonias. The correlation analysis results showed that some textural features were positively correlated with WBC, and NE, and also negatively related to SPO2H and NE. Our results showed that radiomic features can classify COVID-19 patients and other pneumonias patients. The SVM model can achieve an excellent diagnosis of COVID-19.
我们提出了一种利用 CT 胸部图像的放射组学特征来识别 2019 年冠状病毒病（COVID-19）和其他肺炎患者的分类方法。收集了两组参与者的胸部 CT 图像（90 例经 RT-PCR 核酸检测确诊为阳性的 COVID-19 患者和 90 例其他肺炎患者），并手动勾勒出肺炎的感兴趣区域（ROI）。利用放射组学方法提取 ROI 的纹理特征和直方图特征，并从每个样本中获取放射组学特征向量。然后，我们将数据分为两个独立的放射组学队列进行训练（70 例 COVID-19 患者和 70 例其他肺炎患者）和验证（20 例 COVID-19 患者和 20 例其他肺炎患者），并采用支持向量机（SVM）进行分析。该模型使用 20 轮 10 折交叉验证进行训练。最后，在独立验证队列上对最终模型进行单次测试。在 COVID-19 患者中，还进行了相关性分析（多重比较校正- Bonferroni 校正，P<0.05/7），以确定纹理和直方图特征是否与血液的实验室检查指标（如血氧、白细胞、淋巴细胞、中性粒细胞、C 反应蛋白、超敏 C 反应蛋白和红细胞沉降率）相关。最终模型在独立验证队列上具有良好的判别能力，准确率为 89.83%，敏感度为 94.22%，特异度为 85.44%，AUC 为 0.940。这证明放射组学特征具有高度的可区分性，该 SVM 模型可以有效地识别和诊断 COVID-19 和其他肺炎患者。相关性分析结果表明，一些纹理特征与 WBC 和 NE 呈正相关，与 SPO2H 和 NE 呈负相关。我们的结果表明，放射组学特征可以对 COVID-19 患者和其他肺炎患者进行分类。SVM 模型可以实现对 COVID-19 的出色诊断。
