ABSTRACT
Objective: To assess the performance of a novel deep learning (DL)-based artificial intelligence (AI) system in classifying computed tomography (CT) scans of pneumonia patients into different groups, as well as to present an effective clinically relevant machine learning (ML) system based on medical image identification and clinical feature interpretation to assist radiologists in triage and diagnosis. Methods: The 3,463 CT images of pneumonia used in this multi-center retrospective study were divided into four categories: bacterial pneumonia (n = 507), fungal pneumonia (n = 126), common viral pneumonia (n = 777), and COVID-19 (n = 2,053). We used DL methods based on images to distinguish pulmonary infections. A machine learning (ML) model for risk interpretation was developed using key imaging (learned from the DL methods) and clinical features. The algorithms were evaluated using the areas under the receiver operating characteristic curves (AUCs). Results: The median AUC of DL models for differentiating pulmonary infection was 99.5% (COVID-19), 98.6% (viral pneumonia), 98.4% (bacterial pneumonia), 99.1% (fungal pneumonia), respectively. By combining chest CT results and clinical symptoms, the ML model performed well, with an AUC of 99.7% for SARS-CoV-2, 99.4% for common virus, 98.9% for bacteria, and 99.6% for fungus. Regarding clinical features interpreting, the model revealed distinctive CT characteristics associated with specific pneumonia: in COVID-19, ground-glass opacity (GGO) [92.5%; odds ratio (OR), 1.76; 95% confidence interval (CI): 1.71-1.86]; larger lesions in the right upper lung (75.0%; OR, 1.12; 95% CI: 1.03-1.25) with viral pneumonia; older age (57.0 years ± 14.2, OR, 1.84; 95% CI: 1.73-1.99) with bacterial pneumonia; and consolidation (95.8%, OR, 1.29; 95% CI: 1.05-1.40) with fungal pneumonia. Conclusion: For classifying common types of pneumonia and assessing the influential factors for triage, our AI system has shown promising results. Our ultimate goal is to assist clinicians in making quick and accurate diagnoses, resulting in the potential for early therapeutic intervention.
ABSTRACT
Automatic severity assessment and progression prediction can facilitate admission, triage, and referral of COVID-19 patients. This study aims to explore the potential use of lung lesion features in the management of COVID-19, based on the assumption that lesion features may carry important diagnostic and prognostic information for quantifying infection severity and forecasting disease progression. A novel LesionEncoder framework is proposed to detect lesions in chest CT scans and to encode lesion features for automatic severity assessment and progression prediction. The LesionEncoder framework consists of a U-Net module for detecting lesions and extracting features from individual CT slices, and a recurrent neural network (RNN) module for learning the relationship between feature vectors and collectively classifying the sequence of feature vectors. Chest CT scans of two cohorts of COVID-19 patients from two hospitals in China were used for training and testing the proposed framework. When applied to assessing severity, this framework outperformed baseline methods achieving a sensitivity of 0.818, specificity of 0.952, accuracy of 0.940, and AUC of 0.903. It also outperformed the other tested methods in disease progression prediction with a sensitivity of 0.667, specificity of 0.838, accuracy of 0.829, and AUC of 0.736. The LesionEncoder framework demonstrates a strong potential for clinical application in current COVID-19 management, particularly in automatic severity assessment of COVID-19 patients. This framework also has a potential for other lesion-focused medical image analyses.
ABSTRACT
Background: Coronary artery calcification (CAC) is an independent risk factor of major adverse cardiovascular events; however, the impact of CAC on in-hospital death and adverse clinical outcomes in patients with coronavirus disease 2019 (COVID-19) remains unclear. Objective: To explore the association between CAC and in-hospital mortality and adverse events in patients with COVID-19. Methods: This multicenter retrospective cohort study enrolled 2067 laboratory-confirmed COVID-19 patients with definitive clinical outcomes (death or discharge) admitted from 22 tertiary hospitals in China between January 3, 2020 and April 2, 2020. Demographic, clinical, laboratory results, chest CT findings, and CAC on admission were collected. The primary outcome was in-hospital death and the secondary outcome was composed of in-hospital death, admission to intensive care unit (ICU), and requiring mechanical ventilation. Multivariable Cox regression analysis and Kaplan-Meier plots were used to explore the association between CAC and in-hospital death and adverse clinical outcomes. Results: The mean age was 50 years (SD,16) and 1097 (53.1%) were male. A total of 177 patients showed high CAC level, and compared with patients with low CAC, these patients were older (mean age: 49 vs. 69 years, P < 0.001) and more likely to be male (52.0% vs. 65.0%, P = 0.001). Comorbidities, including cardiovascular disease (CVD) ([33.3%, 59/177] vs. [4.7%, 89/1890], P < 0.001), presented more often among patients with high CAC, compared with patients with low CAC. As for laboratory results, patients with high CAC had higher rates of increased D-dimer, LDH, as well as CK-MB (all P < 0.05). The mean CT severity score in high CAC group was also higher than low CAC group (12.6 vs. 11.1, P = 0.005). In multivariable Cox regression model, patients with high CAC were at a higher risk of in-hospital death (hazard ratio [HR], 1.731; 95% CI 1.010-2.971, P = 0.046) and adverse clinical outcomes (HR, 1.611; 95% CL 1.087-2.387, P = 0.018). Conclusion: High CAC is a risk factor associated with in-hospital death and adverse clinical outcomes in patients with confirmed COVID-19, which highlights the importance of calcium load testing for hospitalized COVID-19 patients and calls for attention to patients with high CAC. Supplementary Information: The online version contains supplementary material available at 10.1007/s42058-021-00072-4.