A mask-guided attention deep learning model for COVID-19 diagnosis based on an integrated CT scan images database

The global extent of COVID-19 mutations and the consequent depletion of hospital resources highlighted the necessity of effective computer-assisted medical diagnosis. COVID-19 detection mediated by deep learning models can help diagnose this highly contagious disease and lower infectivity and mortality rates. Computed tomography (CT) is the preferred imaging modality for building automatic COVID-19 screening and diagnosis models. It is well-known that the training set size significantly impacts the performance and generalization of deep learning models. However, accessing a large dataset of CT scan images from an emerging disease like COVID-19 is challenging. Therefore, data efficiency becomes a significant factor in choosing a learning model. To this end, we present a multi-task learning approach, namely, a mask-guided attention (MGA) classifier, to improve the generalization and data efficiency of COVID-19 classification on lung CT scan images. The novelty of this method is compensating for the scarcity of data by employing more supervision with lesion masks, increasing the sensitivity of the model to COVID-19 manifestations, and helping both generalization and classification performance. Our proposed model achieves better overall performance than the single-task (without MGA module) baseline and state-of-the-art models, as measured by various popular metrics.

Predictors of the chest CT score in COVID-19 patients: a cross-sectional study.

BACKGROUND: Since the COVID-19 outbreak, pulmonary involvement was one of the most significant concerns in assessing patients. In the current study, we evaluated patient's signs, symptoms, and laboratory data on the first visit to predict the severity of pulmonary involvement and their outcome regarding their initial findings. METHODS: All referred patients to the COVID-19 clinic of a tertiary referral university hospital were evaluated from April to August 2020. Four hundred seventy-eight COVID-19 patients with positive real-time reverse-transcriptase-polymerase chain reaction (RT-PCR) or highly suggestive symptoms with computed tomography (CT) imaging results with typical findings of COVID-19 were enrolled in the study. The clinical features, initial laboratory, CT findings, and short-term outcomes (ICU admission, mortality, length of hospitalization, and recovery time) were recorded. In addition, the severity of pulmonary involvement was assessed using a semi-quantitative scoring system (0-25). RESULTS: Among 478 participants in this study, 353 (73.6%) were admitted to the hospital, and 42 (8.7%) patients were admitted to the ICU. Myalgia (60.4%), fever (59.4%), and dyspnea (57.9%) were the most common symptoms of participants at the first visit. A review of chest CT scans showed that Ground Glass Opacity (GGO) (58.5%) and consolidation (20.7%) were the most patterns of lung lesions. Among initial clinical and laboratory findings, anosmia (P = 0.01), respiratory rate (RR) with a cut point of 25 (P = 0.001), C-reactive protein (CRP) with a cut point of 90 (P = 0.002), white Blood Cell (WBC) with a cut point of 10,000 (P = 0.009), and SpO2 with a cut point of 93 (P = 0.04) was associated with higher chest CT score. Lung involvement and consolidation lesions on chest CT scans were also associated with a more extended hospitalization and recovery period. CONCLUSIONS: Initial assessment of COVID-19 patients, including symptoms, vital signs, and routine laboratory tests, can predict the severity of lung involvement and unfavorable outcomes.

CT-scan findings of COVID-19 pneumonia based on the time elapsed from the beginning of symptoms to the CT imaging evaluation: a descriptive study in Iran.

Background. Coronavirus disease 2019 (COVID-19) was initially detected in Wuhan city, China. Chest CT features of COVID-19 pneumonia have been investigated mostly in China, and there is very little information available on the radiological findings occurring in other populations. In this study, we aimed to describe the characteristics of chest CT findings in confirmed cases of COVID-19 pneumonia in an Iranian population, based on a time classification.Methods. Eighty-nine patients with COVID-19 pneumonia, confirmed by a real-time RT-PCR test, who were admitted to non-ICU wards and underwent a chest CT scan were retrospectively enrolled. Descriptive evaluation of radiologic findings was performed using a classification based on the time interval between the initiation of the symptoms and chest CT-scan.Results. The median age of patients was 58.0 years, and the median time interval from the onset of symptoms to CT scan evaluation was 7 days. Most patients had bilateral (94.4%) and multifocal (91.0%) lung involvement with peripheral distribution (60.7%). Also, most patients showed involvement of all five lobes (77.5%). Ground-glass opacities (GGO) (84.3%) and mixed GGO with consolidation (80.9%) were the most common identified patterns. We also found that as the time interval between symptoms and CT scan evaluation increased, the predominant pattern changed from GGO to mixed pattern and then to elongated-containing and band-like-opacities-containing pattern; on the other hand, the percentage of lung involvement increased.Conclusions. Bilateral multifocal GGO, and mixed GGO with consolidation were the most common patterns of COVID-19 pneumonia in our study. However, these patterns might change according to the time interval from symptoms.