Prediction of oxygen supplementation by a deep-learning model integrating clinical parameters and chest CT images in COVID-19

Abstract

Purpose: As of March 2023, the number of patients with COVID-19 worldwide is declining, but the early diagnosis of patients requiring inpatient treatment and the appropriate allocation of limited healthcare resources remain unresolved issues. In this study we constructed a deep-learning (DL) model to predict the need for oxygen supplementation using clinical information and chest CT images of patients with COVID-19.

Materials and methods: We retrospectively enrolled 738 patients with COVID-19 for whom clinical information (patient background, clinical symptoms, and blood test findings) was available and chest CT imaging was performed. The initial data set was divided into 591 training and 147 evaluation data. We developed a DL model that predicted oxygen supplementation by integrating clinical information and CT images. The model was validated at two other facilities (n = 191 and n = 230). In addition, the importance of clinical information for prediction was assessed.

Results: The proposed DL model showed an area under the curve (AUC) of 89.9% for predicting oxygen supplementation. Validation from the two other facilities showed an AUC > 80%. With respect to interpretation of the model, the contribution of dyspnea and the lactate dehydrogenase level was higher in the model.

Conclusions: The DL model integrating clinical information and chest CT images had high predictive accuracy. DL-based prediction of disease severity might be helpful in the clinical management of patients with COVID-19.

Keywords: COVID-19; Chest CT images; Deep learning; Disease severity prediction; Explainable AI.

Fig. 1

Development of the DL models. Notes: The first DL model is the clinical network architecture, which is a DL model using clinical information (a). The clinical information (patient background, symptoms, and blood test findings; n = 62 items) was reformed into 62 channels, convo-transposed twice, and passed through a fully connected layer to generate outputs (1 for oxygen supplementation; 0 for no oxygen supplementation). The second DL model is the image network architecture, referring to a previous report based on Densenet [22]. This model was implemented using chest CT images (b). After passing through the convolution layer, the model passed through three transitions: dense block, convolution layer, and average pooling. Then, after going through global average pooling, the model passed through the fully connected layer to produce the output. In each dense block, each network layer has a tightly coupled structure consisting of a 3*3 convolutional layer and a 3*3*3 convolutional layer. These layers are N-connected and have a residual structure where the outputs of each layer are added together from behind

Fig. 2

Development of the proposed DL model and proposed network architecture. Notes: The third model combined the clinical network with the image network. The DL model is the proposed network architecture. The products from the clinical network and image network were combined and passed through a fully connected layer, then through Resnet structures, and finally through a fully connected layer. Then, the final output was generated

Fig. 3

Receiver operating characteristic (ROC) curve analysis of oxygen supplementation prediction by the DL model combined with clinical information and CT images. Notes: The prediction accuracy of the model combining clinical information and CT images was higher than that of clinical information and CT images alone. Green line; Clinical information. Thin blue line; CT image. Indigo line; Total data (Clinical information + CT image)

Fig. 4

Receiver operating characteristic (ROC) curve analysis of oxygen supplementation prediction by the DL model combined with clinical information and CT images using externally-validated data. Notes: The validation results at the other two sites were > 80%. a and b First external validation. c and d Second external validation. Green line; Clinical information. Thin blue line; CT image. Indigo line; Total data (Clinical information + CT image)

Fig. 5

Analysis of factors affecting oxygen supplementation prediction. Notes: Using the learned parameters, the importance of each item was evaluated by the proposed formula. Dyspnea, a clinical symptom, and LDH, a blood test finding, were shown to have a strong influence on the presence of oxygen supplementation. a The contribution of the items among patient background and clinical symptoms. b The contribution of the items among blood test findings. Abbreviations: BMI body mass index; COPD chronic obstructive pulmonary disease; TP total protein; ALB albumin; AG ratio albumin:globulin ratio; AST aspartate aminotransferase; ALT alanine aminotransferase; LDH lactate dehydrogenase; T-Bil total bilirubin; γ-GTP γ-glutamyltransferase; BUN blood urea nitrogen; Cre creatinine; UA uric acid; eGFR estimated glomerular filtration rate; Na sodium; K potassium; Cl chloride ion; CPK creatine phosphokinase; CRP C-reactive protein; GLU glucose; WBC white blood cell; RBC red blood cell; HGB hemoglobin; Hct hematocrit; MCV mean corpuscular volume; MCH mean corpuscular hemoglobin; MCHC mean corpuscular hemoglobin concentration; PLT platelet; Baso basophil; Eosino eosinophil; Neutro neutrophil; Lympho lymphocyte; Mono monocyte; PNI prognostic nutritional index