Our study constructed nomograms based on radiomics and clinical factors for predicting vaccination status in patients with novel coronavirus disease. The results showed that CT score and RadScore were independent factors for predicting all vaccination status. Radiomics is mainly based on radiomics features of the degree of lung inflammation, and nomograms that combine clinical factors can well identify patient vaccination status. At present, some scholars have developed predictive models for evaluating patients with new coronary pneumonia, but the validation of the models needs to be further explored. Therefore, we adopted an internal validation approach of the model, which has been demonstrated .
In the unvaccinated group, compared with the partially vaccinated group and the full vaccination group, the age was significantly older and the peripheral blood lymphocyte count was lower. Studies have shown that the changes of lymphocytes are closely related to the pathogenesis of the virus . SARS-CoV-2 infection can destroy lymph nodes by inhibiting bone marrow or directly inducing immunity, resulting in peripheral blood lymphocytopenia . In our article, the lymphocyte count of the people who were not vaccinated with novel coronavirus vaccine decreased significantly, which suggested that compared with the people who had been vaccinated with novel coronavirus vaccine, the delta variant infection in this group may have more serious immune cell consumption and cellular immune function impairment. Moreover, it is reported that the decrease of peripheral blood lymphocyte count in the early stage of admission is one of the potential early clinical early warning indicators of COVID-19 's severe / critical tendency . We also found that the lymphocyte count in the partial vaccination group was lower than that in the complete vaccination group (1.55 ± 0.70 vs. 1.87 ± 0.70, p < 0.05), which indicated that the lymphocyte count might be valuable in reflecting the severity of coronavirus pneumonia.
From the overall chest CT score analysis, the degree of lung infection was related to the degree of vaccination and the difference was statistically significant. This result further demonstrated the previous findings that booster injection on the basis of vaccination could increase the protective effect of COVID-19 vaccine in humans [25, 26]. Besides, we also conducted a preliminary study on the chest HRCT manifestations of patients infected with unvaccinated, partially vaccinated and completely inoculated delta mutants, revealing some major manifestations on chest CT images. First of all, whether vaccinated or not, exudative changes with or without local consolidation foci were common in positive chest CT, that is, multiple or single ground glass shadows could be accompanied by solid lesions, simple single shadows foci were rare, and thickened vascular bundles could be seen in some shadows. Secondly, the lesions in the lung can involve each lung lobe, but usually the lower lobes of both lungs are relatively common, and the lesions are mostly distributed in the periphery and / or subpleural of the lung, and simple central lesions are rarely seen, which may be related to the pathological mechanism of viral pneumonia invading the pulmonary parenchyma. For example, in the early stage, the lesions are easy to involve the terminal bronchioles and the parenchyma around the respiratory bronchioles, and then spread along the bronchovesicular bundle to the middle lung field . Other signs included interstitial changes, HRCT showed fine reticular linear high-density shadows or stripe focus, and some of them showed "paving stone" changes. Finally, extrapulmonary manifestations of HRCT, such as pleural effusion or mediastinal enlarged lymph nodes, were rare in the three groups.
In this study, 3D-ROI image features extracted by radiomics technology can provide more image information than conventional 2D-ROI, because 3D-ROI can provide more complete features of infected lung volume, easier to capture heterogeneous information of lesions, and eliminate the influence of manual sketching, increasing the repeatability and reliability of radiomics features, so it has higher diagnostic efficiency .
After feature standardization, elimination of redundant features and LASSO feature selection, the radiological score RadScore was constructed, and the chest CT of patients infected with delta mutant was quantitatively analyzed to explore the relationship between RadScore and the vaccination status of patients. The final experimental results show that after feature selection, a total of 10 radiological features constitute RadScore, including 8 texture features and 2 first-order statistical features. Among the 10 radiological features that can predict the vaccination status of patients, 7 imaging features are negatively correlated with RadScore, including those obtained by wavelet filter, mean normalization filter, curvature filter, shot noise filter and Laplace filter post-processing, which can indicate the invasion degree of coronavirus to lung parenchyma. Among them, the larger the eigenvalue of the first-order statistical feature 3D maximum diameter (log_firstorder_log-sigma-1–5-mm-3d-maximum) processed by Laplace filter, indicates that the wider the area of infection, the greater the possibility of patients not being vaccinated, so it is negatively correlated with RadScore. In addition, three imaging features were positively correlated with RadScore, including the texture features processed by gray area size matrix and box filter, which represented the texture features of residual normal lung parenchyma, and their values could indirectly reflect the damage degree of lung parenchyma. We found that there were statistical differences in RadScore among the three groups of patients with different vaccination status. The RadScore of the unvaccinated group was the smallest, the RadScore of the incomplete vaccinated group was the second, and the RadScore of the complete vaccinated group was the largest. The results show that the imaging score RadScore has the ability to identify the vaccination of patients, and it also proves that the larger the RadScore, the greater the possibility of full vaccination, and the smaller the imaging characteristic value showing the size of the infected area in the corresponding RadScore. This result is also consistent with the related literature that COVID-19 vaccine has protective effect on the lungs [29,30,31,32,33].
In addition, our research has some limitations. First of all, retrospective studies may have some selection bias; secondly, the case comes from a single center, the sample size in each group is relatively small, and due to the limited clinical data, more blood biochemical indexes, such as viral load and neutralizing antibody titer, are not included, which is obviously an important issue to be explored in the future. Besides, we exclude patients with basic lung diseases and other basic diseases, and we need to further explore the impact of different basic diseases on COVID-19 vaccine in the future. Finally, the clinical symptoms of the patients included were not dynamically monitored and evaluated. Therefore, further prospective studies with large-scale, multicenter and longer observation periods are needed.