Neutrophil Elastase Inhibitor (Sivelestat) in the Treatment of Acute Respiratory Distress Syndrome Induced by COVID- 19: A Multicenter Retrospective Cohort Study (preprint)

Background Recent studies suggest that neutrophil elastase inhibitor (Sivelestat) may improve pulmonary function and reduce mortality in patients with acute respiratory distress syndrome. We examined the association between receipt of sivelestat and improvement in oxygenation among patients with acute respiratory distress syndrome (ARDS) induced by COVID-19.Methods A large multicentre cohort study of patients with ARDS induced by COVID-19 who had been admitted to intensive care units (ICUs). We used propensity score matching to compare the outcomes of patients treated with sivelestat to those who were not. The differences in continuous outcomes were assessed with the Wilcoxon signed-rank test. Kaplan-Meier method was used to show the 28-day survival curves in the matched cohorts. A log-rank P-test stratified on the matched pairs was used to test the equality of the estimated survival curves. A Cox proportional hazards model that incorporated a robust sandwich-type variance estimator to account for the matched nature of the data was used to estimate hazard ratios (HR). All statistical analyses were performed with SPSS 26.0 and R 4.2.3. A two-sided p-value of < 0.05 was considered statistically significant.Results A total of 387 patients met inclusion criteria, including 259 patients (66.9%) who were treated with sivelestat. In 158 patients matched on the propensity for treatment, receipt of sivelestat was associated with improved oxygenation, decreased Murray lung injury score, increased non-mechanical ventilation time within 28 days, increased alive and ICU-free days within 28 days (HR, 1.85; 95% CI, 1.29 to 2.64; log-rank p < 0.001), shortened ICU stay and ultimately improved survival (HR, 2.78; 95% CI, 1.32 to 5.88; log-rank p = 0.0074).Conclusions Among patients with ARDS induce by COVID-19, sivelestat administration is associated with improved clinical outcomes.

The Covid Response Study (COVRES) Protocol: a comprehensive genomic, immunological, proteomic and phenotypic comparison of hospitalised versus non-hospitalised SARS-CoV-2 patients in Northern Ireland (preprint)

Objective  The Covid Response Study (COVRES, NCT05548829) aims to carry out an integrated multi-omic analysis of factors contributing to host susceptibility to SARS-CoV-2 among a patient cohort of 1000 people from the geographically isolated island of Ireland. Background  Health organisations and countries around the world have found it difficult to control the spread of the coronavirus disease 2019. To minimise the impact on the NHS and improve patient care, there is a drive for rapid tests capable of detecting individuals who are at high risk of contracting severe COVID-19. Early work focused on single omic approaches, highlighting a limited amount of information. Study Design The protocol below describes the study to be carried out in Northern Ireland (NI-COVRES) by Ulster University, the Republic of Ireland component will be described separately. All participants (n = 519) were recruited from the Western Health and Social Care Trust, Northern Ireland, forty patients are also being followed up at 1, 3, 6 and 12 months to assess the longitudinal impact of infection on symptoms, general health, and immune response, this is ongoing. Methods Data will be sourced from whole blood, saliva samples, and clinical data from the Northern Ireland Electronic Care Record, general health questionnaire, and the GHQ12 mental health survey. Saliva and blood samples were processed for DNA and RNA prior to whole genomic sequencing, RNA sequencing, DNA methylation, microbiome, 16S, and proteomic analysis. Multi-omics data will be combined with clinical data to produce sensitive and specific prognostic models of severity risk. Results An initial profile of the cohort has been completed: n = 249 hospitalised and n = 270 non-hospitalised patients were recruited, 64% were female, the mean age was 45 years. High levels of comorbidity were evident in the hospitalised cohort, with cardiovascular disease and metabolic and respiratory disorders (P < 0.001) being the most significant. Conclusion This study will provide a comprehensive opportunity to study multi-omic mechanisms of COVID-19 severity in re-contactable participants. Trial Registration - The trial has been registered as an observational study on clinicaltrials.gov as NCT05548829. An outline of the trial protocol is included; SPIRIT checklist (Supplementary Fig. 1).

Different effects of vaccine on VST in critical and non-critical COVID-19 patients: A retrospective study of 363 cases.

Aim: To explore the risk factors of prolonged viral shedding time (VST) in critical/non-critical COVID-19 patients during hospitalization. Methods: In this retrospective study, we enrolled 363 patients with SARS-CoV-2 infection admitted in a designated hospital during the COVID-19 outbreak in Nanjing Lukou International Airport. Patients were divided into critical (n = 54) and non-critical (n = 309) groups. We analyzed the relationship between the VST and demographics, clinical characteristics, medications, and vaccination histories, respectively. Results: The median duration of VST was 24 d (IQR, 20-29) of all patients. The VST of critical cases was longer than non-critical cases (27 d, IQR, 22.0-30.0 vs. 23 d, IQR 20-28, P < 0.05). Cox proportional hazards model showed that ALT (HR = 1.610, 95%CI 1.186-2.184, P = 0.002) and EO% (HR = 1.276, 95%CI 1.042-1.563, P = 0.018) were independent factors of prolonged VST in total cases; HGB (HR = 0.343, 95%CI 0.162-0.728, P = 0.005) and ALP (HR = 0.358, 95%CI 0.133-0.968, P = 0.043) were independent factors of prolonged VST in critical cases, while EO% (HR = 1.251, 95%CI 1.015-1.541, P = 0.036) was the independent factor of prolonged VST in non-critical cases. Vaccinated critical cases showed higher levels of SARS-CoV-2-IgG (1.725 S/CO, IQR 0.3975-28.7925 vs 0.07 S/CO, IQR 0.05-0.16, P < 0.001) and longer VSTs (32.5 d, IQR 20.0-35.25 vs 23 d, IQR 18.0-30.0, P = 0.011) compared with unvaccinated critical patients. Fully vaccinated non-critical cases, however, presented higher levels of SARS-CoV-2-IgG (8.09 S/CO, IQR 1.6975-55.7825 vs 0.13 S/CO IQR 0.06-0.41, P < 0.001) and shorter VSTs (21 d, IQR 19.0-28.0 vs 24 d, IQR 21.0-28.5, P = 0.013) compared with unvaccinated non-critical patients. Conclusions: Our results suggested that risk factors of prolonged VST were different between critical and non-critical COVID-19 patients. Increased level of SARS-CoV-2-IgG and vaccination did not shorten the VST and hospital stay in critical COVID-19 patients.

Boosting Convolution with Efficient MLP-Permutation for Volumetric Medical Image Segmentation (preprint)

Recently, the advent of vision Transformer (ViT) has brought substantial advancements in 3D dataset benchmarks, particularly in 3D volumetric medical image segmentation (Vol-MedSeg). Concurrently, multi-layer perceptron (MLP) network has regained popularity among researchers due to their comparable results to ViT, albeit with the exclusion of the resource-intensive self-attention module. In this work, we propose a novel permutable hybrid network for Vol-MedSeg, named PHNet, which capitalizes on the strengths of both convolution neural networks (CNNs) and MLP. PHNet addresses the intrinsic isotropy problem of 3D volumetric data by employing a combination of 2D and 3D CNNs to extract local features. Besides, we propose an efficient multi-layer permute perceptron (MLPP) module that captures long-range dependence while preserving positional information. This is achieved through an axis decomposition operation that permutes the input tensor along different axes, thereby enabling the separate encoding of the positional information. Furthermore, MLPP tackles the resolution sensitivity issue of MLP in Vol-MedSeg with a token segmentation operation, which divides the feature into smaller tokens and processes them individually. Extensive experimental results validate that PHNet outperforms the state-of-the-art methods with lower computational costs on the widely-used yet challenging COVID-19-20 and Synapse benchmarks. The ablation study also demonstrates the effectiveness of PHNet in harnessing the strengths of both CNNs and MLP.

PSTCNN: Explainable COVID-19 diagnosis using PSO-guided self-tuning CNN

Since 2019, the coronavirus disease-19 (COVID-19) has been spreading rapidly worldwide, posing an unignorable threat to the global economy and human health. It is a disease caused by severe acute respiratory syndrome coronavirus 2, a single-stranded RNA virus of the genus Betacoronavirus. This virus is highly infectious and relies on its angiotensin-converting enzyme 2-receptor to enter cells. With the increase in the number of confirmed COVID-19 diagnoses, the difficulty of diagnosis due to the lack of global healthcare resources becomes increasingly apparent. Deep learning-based computer-aided diagnosis models with high generalisability can effectively alleviate this pressure. Hyperparameter tuning is essential in training such models and significantly impacts their final performance and training speed. However, traditional hyperparameter tuning methods are usually time-consuming and unstable. To solve this issue, we introduce Particle Swarm Optimisation to build a PSO-guided Self-Tuning Convolution Neural Network (PSTCNN), allowing the model to tune hyperparameters automatically. Therefore, the proposed approach can reduce human involvement. Also, the optimisation algorithm can select the combination of hyperparameters in a targeted manner, thus stably achieving a solution closer to the global optimum. Experimentally, the PSTCNN can obtain quite excellent results, with a sensitivity of 93.65% ± 1.86%, a specificity of 94.32% ± 2.07%, a precision of 94.30% ± 2.04%, an accuracy of 93.99% ± 1.78%, an F1-score of 93.97% ± 1.78%, Matthews Correlation Coefficient of 87.99% ± 3.56%, and Fowlkes-Mallows Index of 93.97% ± 1.78%. Our experiments demonstrate that compared to traditional methods, hyperparameter tuning of the model using an optimisation algorithm is faster and more effective. [ FROM AUTHOR]

Clinical characteristics and analysis of risk factors for disease progression of patients with SARS-CoV-2 Omicron variant infection: A retrospective study of 25207 cases in a Fangcang hospital.

Objectives: To summarize the clinical characteristics of patients infected by SARS-CoV-2 omicron variant and explore the risk factors affecting the progression in a Fangcang hospital, Shanghai, China. Methods: A total of 25,207 patients were retrospectively enrolled. We described the clinical characteristics and performed univariate and multivariate logistic regression analysis to identify the risk factors for non-severe to severe COVID-19 or death. Results: According to the outcomes, there were 39 severe patients (including 1 death) and 25,168 non-severe patients enrolled in this study. Among the 25,207 cases, the median age was 45 years (IQR 33-54), and 65% patients were male. Cough (44.5%) and expectoration (38.4%) were the most two common symptoms. Hypertension (10.4%) and diabetes (3.5%) were most two common comorbidities. Most patients (81.1%) were fully vaccinated. The unvaccinated and partially vaccinated patients were 15.1% and 3.9%, respectively. The length of viral shedding time was six days (IQR 4-9) in non-severe patients. Multivariate logistic regression analysis suggested that age (OR=1.062, 95%CI 1.034-1.090, p<0.001), fever (OR=2.603, 95%CI 1.061-6.384, p=0.037), cough (OR=0.276, 95%CI 0.119-0.637, p=0.003), fatigue (OR=4.677, 95%CI 1.976-11.068, p<0.001), taste disorders (OR=14.917, 95%CI 1.884-118.095, p=0.010), and comorbidity (OR=2.134, 95%CI 1.059-4.302, p=0.034) were predictive factors for deterioration of SARS-CoV-2 omicron variant infection. Conclusions: Non-critical patients have different clinical characteristics from critical patients. Age, fever, cough, fatigue, taste disorders, and comorbidity are predictors for the deterioration of SARS-CoV-2 omicron variant infection.

Mutagenesis and structural studies reveal the basis for the specific binding of SARS-CoV-2 SL3 RNA element with human TIA1 protein (preprint)

Viral RNA-host protein interactions are indispensable during RNA virus transcription and replication. However, the detailed structural and dynamical features of the interactions between viral RNA and various host proteins remain largely elusive. Here, we characterized the binding interface for the SARS-CoV-2 stem-loop 3 (SL3) cis-acting element to human TIA1 protein with a combined theoretical and experimental approach including molecular modeling, free energy calculations, and electrophoretic mobility shift assays (EMSA). As a highly structured and conserved cis-acting element, SARS-CoV-2 SL3 RNA element was found to have a high binding affinity (Kd ~ 780 nM) to human TIA1 protein, with its hairpin and 3’-terminal loops playing essential roles in a sequence-dependent manner. Our molecular dynamics simulations revealed that the aromatic stacking, specific hydrogen bonds, and hydrophobic interactions collectively direct the specific binding of SL3 RNA element to TIA1, in which notable conformation changes both in protein domain arrangement and RNA 3D structure adaptation were observed. Further evaluations of in silico mutagenesis predictions with electrophoretic assays validated our proposed 3D binding model and also revealed SL3 A68U variant has an enhanced binding affinity (~ 1.7-fold) to TIA1 protein than the wild type. Finally, we found that the human TIA1 protein could interact with conserved SL3 RNA elements within other betacoronavirus lineages as well. These findings open a new avenue to explore the viral RNA-host protein interactions for SARS-CoV-2 infection and provide a pioneering structural basis for novel RNA-targeting antiviral drug design.

Randomized, placebo-controlled, single-blind phase 1 studies of the safety, tolerability, and pharmacokinetics of BRII-196 and BRII-198, SARS-CoV-2 spike-targeting monoclonal antibodies with an extended half-life in healthy adults

Background: BRII-196 and BRII-198 are two anti-SARS-CoV-2 monoclonal neutralizing antibodies as a cocktail therapy for treating COVID-19 with a modified Fc region that extends half-life. Methods: Safety, tolerability, pharmacokinetics, and immunogenicity of BRII-196 and BRII-198 were investigated in first-in-human, placebo-controlled, single ascending dose phase 1 studies in healthy adults. 44 participants received a single intravenous infusion of single BRII-196 or BRII-198 up to 3,000 mg, or BRII-196 and BRII-198 combination up to 1500/1500 mg, or placebo and were followed up for 180 days. Primary endpoints were incidence of adverse events (AEs) and changes from pre-dose baseline in clinical assessments. Secondary endpoints included pharmacokinetics profiles of BRII-196/BRII-198 and detection of anti-drug antibodies (ADAs). Plasma neutralization activities against SARS-CoV-2 Delta live virus in comparison to post-vaccination plasma were evaluated as exploratory endpoints. Results: All infusions were well-tolerated without systemic or local infusion reactions, dose-limiting AEs, serious AEs, or deaths. Most treatment-emergent AEs were isolated asymptomatic laboratory abnormalities of grade 1-2 in severity. BRII-196 and BRII-198 displayed pharmacokinetics characteristic of Fc-engineered human IgG1 with mean terminal half-lives of 44.6–48.6 days and 72.2–83.0 days, respectively, with no evidence of interaction or significant anti-drug antibody development. Neutralizing activities against the live virus of the SARS-CoV-2 Delta variant were maintained in plasma samples taken on day 180 post-infusion. Conclusion: BRII-196 and BRII-198 are safe, well-tolerated, and suitable therapeutic or prophylactic options for SARS-CoV-2 infection. Clinical Trial Registration:ClinicalTrials.gov under identifiers NCT04479631, NCT04479644, and NCT04691180.

Correlation of CT-derived pectoralis muscle status and COVID-19 induced lung injury in elderly patients.

OBJECTIVES: To explore the association between CT-derived pectoralis muscle index (PMI) and COVID-19 induced lung injury. METHODS: We enrolled 116 elderly COVID-19 patients linked to the COVID-19 outbreak in Nanjing Lukou international airport. We extracted three sessions of their CT data, including one upon admission (T1), one during the first 2 weeks when lung injury peaked (T2) and one on day 14 ± 2 (T3). Lung injury was assessed by CT severity score (CTSS) and pulmonary opacity score (POS). Pneumonia evolution was evaluated by changes of CT scores at T2 from T1(Δ). RESULTS: The maximum CT scores in low PMI patients were higher than those of normal PMI patients, including CTSS1 (7, IQR 6-10 vs. 5, IQR 3-6, p < 0.001), CTSS2 (8, IQR 7-11 vs. 5, IQR 4-7, p < 0.001) and POS (2, IQR 1-2.5 vs. 1, IQR 1-2, p < 0.001). Comorbidity (OR = 6.15, p = 0.023) and the presence of low PMI (OR = 5.43, p = 0.001) were predictors of lung injury aggravation with ΔCTSS1 > 4. The presence of low PMI (OR = 5.98, p < 0.001) was the predictor of lung injury aggravation with ΔCTSS2 > 4. Meanwhile, presence of low PMI (OR = 2.82, p = 0.042) and incrementally increasing D-dimer (OR = 0.088, p = 0.024) were predictors of lung injury aggravation with ΔPOS = 2. CONCLUSIONS: PMI can be easily assessed on chest CT images and can potentially be used as one of the markers to predict the severity of lung injury in elderly COVID-19 patients.

A broad-spectrum antiviral fusion protein, Pheromonicin demonstrated protective efficacy against SARS-CoV-2 variants in vitro and in vivo models (preprint)

The unprecedented coronavirus disease (COVID-19) epidemic has created a worldwide public health emergency, and there is an urgent need to develop an effective antiviral drug to control this severe infectious disease. Here, we found that the E, or M membrane proteins of coronavirus could be targeted by a 28-residue antibody mimetic by fusing two antibody Fab complementarity-determining regions (VHCDR1 and VLCDR3) through a cognate framework region (VHFR2) of the antibodies which recognize the coronavirus E or M proteins. We constructed a fusion protein, pheromonicin-covid-19 (PMC-covid-19), by linking colicin Ia, a bactericidal molecule produced by E.coli which kills target cells by forming a voltage-dependent channel in target lipid bilayers, to that antibody mimetic. The E, or M protein/antibody mimetic interaction initiated the formation of irreversible PMC-covid-19 channel in the covid-19 envelope and infected host cell membrane resulting in leakage of cellular contents. PMC-covid-19 demonstrates broad-spectrum protective efficacy against tested variants of coronavirus severe acute respiratory syndrome (p<0.01-0.0001). PMC-covid-19 significantly altered outcomes of in vivo fatal covid-19 challenge infection without evident toxicity, making it an appropriate candidate for further clinical evaluation.

A Survey on Machine Learning in COVID-19 Diagnosis

Since Corona Virus Disease 2019 outbreak, many expert groups worldwide have studied the problem and proposed many diagnostic methods. This paper focuses on the research of Corona Virus Disease 2019 diagnosis. First, the procedure of the diagnosis based on machine learning is introduced in detail, which includes medical data collection, image preprocessing, feature extraction, and image classification. Then, we review seven methods in detail: transfer learning, ensemble learning, unsupervised learning and semi-supervised learning, convolutional neural networks, graph neural networks, explainable deep neural networks, and so on. What’s more, the advantages and limitations of different diagnosis methods are compared. Although the great achievements in medical images classification in recent years, Corona Virus Disease 2019 images classification based on machine learning still encountered many problems. For example, the highly unbalanced dataset, the difficulty of collecting labeled data, and the poor quality of the data. Aiming at these problems, we propose some solutions and provide a comprehensive presentation for future research.

Structure prediction of the druggable fragments in SARS-CoV-2 untranslated regions (preprint)

The outbreak of the COVID-19 pandemic has led to intensive studies of both the structure and replication mechanism of SARS-CoV-2. In spite of some secondary structure experiments being carried out, the 3D structure of the key function regions of the viral RNA has not yet been well understood. At the beginning of COVID-19 breakout, RNA-Puzzles community attempted to envisage the three-dimensional structure of 5'- and 3'-Un-Translated Regions (UTRs) of the SARS-CoV-2 genome. Here, we report the results of this prediction challenge, presenting the methodologies developed by six participating groups and discussing 100 RNA 3D models (60 models of 5'-UTR and 40 of 3'-UTR) predicted through applying both human experts and automated server approaches. We describe the original protocol for the reference-free comparative analysis of RNA 3D structures designed especially for this challenge. We elaborate on the deduced consensus structure and the reliability of the predicted structural motifs. All the computationally simulated models, as well as the development and the testing of computational tools dedicated to 3D structure analysis, are available for further study.

Novel Coronavirus SARS-CoV-2 case analysis and typical cases

To provide scientific theoretical reference for diagnosis and treatment of patients with SARS-CoV-2 infection, this study collected typical COVID-19 patients' pathological reports and analyzed the pathological characteristics of the important systems and parts of human body, including respiratory system, circulatory system, digestive system, genitourinary system, nervous system and eye. Cytokine release syndrome(CRS), a common cause of death in COVID-19 patients, was discussed in this research.

Deep Rank-Based Average Pooling Network for Covid-19 Recognition

(Aim) To make a more accurate and precise COVID-19 diagnosis system, this study proposed a novel deep rank-based average pooling network (DRAPNet) model, i.e., deep rank-based average pooling network, for COVID-19 recognition. (Methods) 521 subjects yield 1164 slice images via the slice level selection method. All the 1164 slice images comprise four categories: COVID-19 positive;community-acquired pneumonia;second pulmonary tuberculosis;and healthy control. Our method firstly introduced an improved multiple-way data augmentation. Secondly, an n-conv rank-based average pooling module (NRAPM) was proposed in which rank-based pooling—particularly, rank-based average pooling (RAP)—was employed to avoid overfitting. Third, a novel DRAPNet was proposed based on NRAPM and inspired by the VGG network. Grad-CAM was used to generate heatmaps and gave our AI model an explainable analysis. (Results) Our DRAPNet achieved a micro-averaged F1 score of 95.49% by 10 runs over the test set. The sensitivities of the four classes were 95.44%, 96.07%, 94.41%, and 96.07%, respectively. The precisions of four classes were 96.45%, 95.22%, 95.05%, and 95.28%, respectively. The F1 scores of the four classes were 95.94%, 95.64%, 94.73%, and 95.67%, respectively. Besides, the confusion matrix was given. (Conclusions) The DRAPNet is effective in diagnosing COVID-19 and other chest infectious diseases. The RAP gives better results than four other methods: strided convolution, l2-norm pooling, average pooling, and max pooling.

Stability of SARS-CoV-2 on Cold-chain Food (preprint)

Although close exposure to respiratory droplets from an infected patient is the main transmission route of SARS-CoV-2, touching contaminated surfaces and objects might also contribute to transmission of this virus. There are increasing reports that cold-chain food and food packages have been tested with SARS-CoV-2, raising concerns that importation of contaminated food could be a source for transmission of SARS-CoV-2. The survival of SARS-CoV-2 on cold-chain food was investigated. We found SARS-CoV-2 remained viable for more than three weeks on both beef and mutton at 4°C. Furthermore, at freezing temperature, SARS-CoV-2 can easily survive for more than eight weeks. Our study showed the ability of SARS-CoV-2 to survive for a long time on cold-chain food at low temperatures.

Probabilistic combination of eigenlungs-based classifiers for COVID-19 diagnosis in chest CT images (preprint)

The outbreak of the COVID-19 (Coronavirus disease 2019) pandemic has changed the world. According to the World Health Organization (WHO), there have been more than 100 million confirmed cases of COVID-19, including more than 2.4 million deaths. It is extremely important the early detection of the disease, and the use of medical imaging such as chest X-ray (CXR) and chest Computed Tomography (CCT) have proved to be an excellent solution. However, this process requires clinicians to do it within a manual and time-consuming task, which is not ideal when trying to speed up the diagnosis. In this work, we propose an ensemble classifier based on probabilistic Support Vector Machine (SVM) in order to identify pneumonia patterns while providing information about the reliability of the classification. Specifically, each CCT scan is divided into cubic patches and features contained in each one of them are extracted by applying kernel PCA. The use of base classifiers within an ensemble allows our system to identify the pneumonia patterns regardless of their size or location. Decisions of each individual patch are then combined into a global one according to the reliability of each individual classification: the lower the uncertainty, the higher the contribution. Performance is evaluated in a real scenario, yielding an accuracy of 97.86%. The large performance obtained and the simplicity of the system (use of deep learning in CCT images would result in a huge computational cost) evidence the applicability of our proposal in a real-world environment.

Anal swab as a potentially optimal specimen for SARS-CoV-2 detection to evaluate hospital discharge of COVID-19 patients

Since December 2019, an outbreak of SARS coronavirus 2 (SARS-CoV-2) began in Wuhan, and has rapidly spread worldwide. Previously, discharged patients with coronavirus disease 2019 (COVID-19) patients met the criteria of China's pneumonia diagnosis and treatment program of novel coronavirus infection (trial version 7) for cure of viral infection. Nevertheless, positive detection of SARS-CoV-2 has been found again in several cured COVID-19 patients, leading to conflicts with current criteria. Here, we report clinically cured cases with positive results only in anal swabs, and investigate the clinical value of anal swabs for SARS-CoV-2 detection.

Classification of Positive COVID-19 CT Scans Using Deep Learning

In medical imaging, computer vision researchers are faced with a variety of features for verifying the authenticity of classifiers for an accurate diagnosis. In response to the coronavirus 2019 (COVID-19) pandemic, new testing procedures, medical treatments, and vaccines are being developed rapidly. One potential diagnostic tool is a reverse-transcription polymerase chain reaction (RT-PCR). RT-PCR, typically a time-consuming process, was less sensitive to COVID-19 recognition in the disease’s early stages. Here we introduce an optimized deep learning (DL) scheme to distinguish COVID-19-infected patients from normal patients according to computed tomography (CT) scans. In the proposed method, contrast enhancement is used to improve the quality of the original images. A pretrained DenseNet-201 DL model is then trained using transfer learning. Two fully connected layers and an average pool are used for feature extraction. The extracted deep features are then optimized with a Firefly algorithm to select the most optimal learning features. Fusing the selected features is important to improving the accuracy of the approach;however, it directly affects the computational cost of the technique. In the proposed method, a new parallel high index technique is used to fuse two optimal vectors;the outcome is then passed on to an extreme learning machine for final classification. Experiments were conducted on a collected database of patients using a 70:30 training: Testing ratio. Our results indicated an average classification accuracy of 94.76% with the proposed approach. A comparison of the outcomes to several other DL models demonstrated the effectiveness of our DL method for classifying COVID-19 based on CT scans.

Uncertainty-driven ensembles of deep architectures for multiclass classification. Application to COVID-19 diagnosis in chest X-ray images (preprint)

Respiratory diseases kill million of people each year. Diagnosis of these pathologies is a manual, time-consuming process that has inter and intra-observer variability, delaying diagnosis and treatment. The recent COVID-19 pandemic has demonstrated the need of developing systems to automatize the diagnosis of pneumonia, whilst Convolutional Neural Network (CNNs) have proved to be an excellent option for the automatic classification of medical images. However, given the need of providing a confidence classification in this context it is crucial to quantify the reliability of the model's predictions. In this work, we propose a multi-level ensemble classification system based on a Bayesian Deep Learning approach in order to maximize performance while quantifying the uncertainty of each classification decision. This tool combines the information extracted from different architectures by weighting their results according to the uncertainty of their predictions. Performance of the Bayesian network is evaluated in a real scenario where simultaneously differentiating between four different pathologies: control vs bacterial pneumonia vs viral pneumonia vs COVID-19 pneumonia. A three-level decision tree is employed to divide the 4-class classification into three binary classifications, yielding an accuracy of 98.06% and overcoming the results obtained by recent literature. The reduced preprocessing needed for obtaining this high performance, in addition to the information provided about the reliability of the predictions evidence the applicability of the system to be used as an aid for clinicians.