ObjectiveTo analyze dynamic characteristics of peripheral blood cells in patients with different types of coronavirus disease 2019 (COVID-19), so as to investigate the predictive value of peripheral blood cells and their dynamic changes for clinical outcome of patients with COVID-19.MethodsForty-eight patients with COVID-19 were collected and analyzed from East Hospital of Renmin Hospital of Wuhan University from February 2 to March 15, 2020. These patients were divided into general group (group A, 17 cases), severe survival group (group B, 21 cases), and severe death group (group C, 10 cases). Blood routine examination was done and analyzed before and after admission and among the three groups. The changes of neutrophils and lymphocytes were compared. The predictive power of neutrophils, lymphocytes, neutrophil-to-lymphocyte ratio (NLR), and platelet-to-lymphocyte ratio (PLR) for clinical outcomes was analyzed through the receiver operating characteristic (ROC) curve.ResultsIn group B, the lymphocyte count at discharge was significantly higher than at admission (P=0.002), and the neutrophil count, NLR and PLR were significantly lower than at admission (P values were 0.012, 0.001 and 0.007, respectively). The lymphocyte counts in the A, B, and C groups were ranked from high to low upon admission, and the differences among the three groups were statistically significant (P values were 0.020, <0.001 and 0.006 for the contrasts between groups A and B, groups A and C, groups B and C, respectively), the NLR were ranked from low to high, and the differences among the three groups were statistically significant (P values were 0.001, <0.001 and 0.026 for the contrasts between groups A and B, groups A and C, groups B and C, respectively). Before discharge or death, there was no significant difference in lymphocyte counts and NLR between A and B groups (P>0.05), and there were statistically significant differences between group C and groups A and B (all P values were<0.001). The proportions of “Neutrophils Lymphocytes Convergence” in groups A and B were 64.7% and 76.2%, respectively, which were significantly higher than that in group C (10.0%). The proportions of “Neutrophils Lymphocytes Separation” in group C was 70.0%, which was significantly higher than those in groups A (0) and B (4.8%). The area under the curve of NLR predicting patients with severe disease (excluding death) was 0.843, with the sensitivity and specificity of ≥3.55 be 0.810 and 0.882; The area under the curve of lymphocyte count predicting death in severe patients was 0.845, with the sensitivity and specificity be 0.700 and 0.905, respectively.ConclusionsDynamic changes in the composition of peripheral blood cells are one of the clinical features of COVID-19, “Neutrophils Lymphocytes Convergence” and “Neutrophils Lymphocytes Separation” predict better and worse clinical outcomes, respectively. NLR and lymphocyte counts are effective indicators for predicting the severity and death of COVID-19.
In computer-aided medical diagnosis, obtaining labeled medical image data is expensive, while there is a high demand for model interpretability. However, most deep learning models currently require a large amount of data and lack interpretability. To address these challenges, this paper proposes a novel data augmentation method for medical image segmentation. The uniqueness and advantages of this method lie in the utilization of gradient-weighted class activation mapping to extract data efficient features, which are then fused with the original image. Subsequently, a new channel weight feature extractor is constructed to learn the weights between different channels. This approach achieves non-destructive data augmentation effects, enhancing the model's performance, data efficiency, and interpretability. Applying the method of this paper to the Hyper-Kvasir dataset, the intersection over union (IoU) and Dice of the U-net were improved, respectively; and on the ISIC-Archive dataset, the IoU and Dice of the DeepLabV3+ were also improved respectively. Furthermore, even when the training data is reduced to 70 %, the proposed method can still achieve performance that is 95 % of that achieved with the entire dataset, indicating its good data efficiency. Moreover, the data-efficient features used in the method have interpretable information built-in, which enhances the interpretability of the model. The method has excellent universality, is plug-and-play, applicable to various segmentation methods, and does not require modification of the network structure, thus it is easy to integrate into existing medical image segmentation method, enhancing the convenience of future research and applications.