Learning from pseudo-lesion: a self-supervised framework for COVID-19 diagnosis

The Coronavirus disease 2019 (COVID-19) has rapidly spread all over the world since its first report in December 2019, and thoracic computed tomography (CT) has become one of the main tools for its diagnosis. In recent years, deep learning-based approaches have shown impressive performance in myriad image recognition tasks. However, they usually require a large number of annotated data for training. Inspired by ground glass opacity, a common finding in COIVD-19 patient's CT scans, we proposed in this paper a novel self-supervised pretraining method based on pseudo-lesion generation and restoration for COVID-19 diagnosis. We used Perlin noise, a gradient noise based mathematical model, to generate lesion-like patterns, which were then randomly pasted to the lung regions of normal CT images to generate pseudo-COVID-19 images. The pairs of normal and pseudo-COVID-19 images were then used to train an encoder–decoder architecture-based U-Net for image restoration, which does not require any labeled data. The pretrained encoder was then fine-tuned using labeled data for COVID-19 diagnosis task. Two public COVID-19 diagnosis datasets made up of CT images were employed for evaluation. Comprehensive experimental results demonstrated that the proposed self-supervised learning approach could extract better feature representation for COVID-19 diagnosis, and the accuracy of the proposed method outperformed the supervised model pretrained on large-scale images by 6.57% and 3.03% on SARS-CoV-2 dataset and Jinan COVID-19 dataset, respectively.

Learning from pseudo-lesion: a self-supervised framework for COVID-19 diagnosis.

The Coronavirus disease 2019 (COVID-19) has rapidly spread all over the world since its first report in December 2019, and thoracic computed tomography (CT) has become one of the main tools for its diagnosis. In recent years, deep learning-based approaches have shown impressive performance in myriad image recognition tasks. However, they usually require a large number of annotated data for training. Inspired by ground glass opacity, a common finding in COIVD-19 patient's CT scans, we proposed in this paper a novel self-supervised pretraining method based on pseudo-lesion generation and restoration for COVID-19 diagnosis. We used Perlin noise, a gradient noise based mathematical model, to generate lesion-like patterns, which were then randomly pasted to the lung regions of normal CT images to generate pseudo-COVID-19 images. The pairs of normal and pseudo-COVID-19 images were then used to train an encoder-decoder architecture-based U-Net for image restoration, which does not require any labeled data. The pretrained encoder was then fine-tuned using labeled data for COVID-19 diagnosis task. Two public COVID-19 diagnosis datasets made up of CT images were employed for evaluation. Comprehensive experimental results demonstrated that the proposed self-supervised learning approach could extract better feature representation for COVID-19 diagnosis, and the accuracy of the proposed method outperformed the supervised model pretrained on large-scale images by 6.57% and 3.03% on SARS-CoV-2 dataset and Jinan COVID-19 dataset, respectively.

Introduction to the antibiotic and antiviral compounds themed collection.

Guest editors Michael J. Sofia and Xuechen Li introduce the themed collection on antibiotic and antiviral compounds.

Non-Neutralizing Epitopes Shade Neutralizing Epitopes against Omicron in a Multiple Epitope-Based Vaccine.

The ongoing coronavirus disease 2019 pandemic has raised concerns about the risk of re-infection. Non-neutralizing epitopes are one of the major reasons for antibody-dependent enhancement. Past studies on the ancestral severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) have revealed an infectivity-enhancing site on the ancestral SARS-CoV-2 spike protein. However, infection enhancement associated with the SARS-CoV-2 Omicron strain remains elusive. In this study, we examined the antibodies induced by a multiple epitope-based vaccine, which showed infection enhancement for the Omicron strain but not for the ancestral SARS-CoV-2 or Delta strain. By examining the antibodies induced by single epitope-based vaccines, we identified a conserved epitope, IDf (450-469), with neutralizing activity against ancestral SARS-CoV-2, Delta, and Omicron. Although neutralizing epitopes are present in the multiple epitope-based vaccine, other immunodominant non-neutralizing epitopes such as IDg (480-499) can shade their neutralizing activity, leading to infection enhancement of Omicron. Our study provides up-to-date epitope information on SARS-CoV-2 variants to help design better vaccines or antibody-based therapeutics against future variants.

A novel linker-immunodominant site (LIS) vaccine targeting the SARS-CoV-2 spike protein protects against severe COVID-19 in Syrian hamsters.

The Coronavirus Disease 2019 (COVID-19) pandemic is unlikely to abate until sufficient herd immunity is built up by either natural infection or vaccination. We previously identified ten linear immunodominant sites on the SARS-CoV-2 spike protein of which four are located within the RBD. Therefore, we designed two linkerimmunodominant site (LIS) vaccine candidates which are composed of four immunodominant sites within the RBD (RBD-ID) or all the 10 immunodominant sites within the whole spike (S-ID). They were administered by subcutaneous injection and were tested for immunogenicity and in vivo protective efficacy in a hamster model for COVID-19. We showed that the S-ID vaccine induced significantly better neutralizing antibody response than RBD-ID and alum control. As expected, hamsters vaccinated by S-ID had significantly less body weight loss, lung viral load, and histopathological changes of pneumonia. The S-ID has the potential to be an effective vaccine for protection against COVID-19.