A radiomics based approach using adrenal gland and periadrenal fat CT images to allocate COVID-19 health care resources fairly

Figures

• 

  Fig. 1. The workflow of the study

• 

  Fig. 2. The framework of the used V-net

• 

  Fig. 3. The comparison of auto-segmentation results and ground truth on representative cases from the training set. The upper line shows the ground truth, and the lower line shows the auto-segmentation results. The green color shows the right adrenal glands, whereas red denotes the left. Yellow shows right periadrenal fat, and purple shows left

• 

  Fig. 4. Correlation heatmap in training, validation, and test set. FM = fusion of adrenal gland and periadrenal fat model; CM = clinical model; RN = radiomics nomogram

• 

  Fig. 5. ROC curves in training, validation, and test sets. AM = adrenal gland model; PM = periadrenal fat model; FM = fusion of adrenal gland and periadrenal fat model; CM = clinical model; RN = radiomics nomogram

• 

  Fig. 6. Box plots summarizing rad score and 7 clinical indicators between moderate and severe patient sets in training, validation, and test sets. In each box plot, the horizontal line crossing the box is the median, and the bottom and top are the lower and upper quartiles. FM = fusion of adrenal gland and periadrenal fat model; CM = clinical model; RN = radiomics nomogram; LDH = lactate dehydrogenase; L = lymphocyte count; Hb = hemoglobin, DD = D dimer; WBC = white blood cells count; TT = thrombin time and TP = total protein

• 

  Fig. 7. Radiomics nomogram developed in training set with radiomics features, lactate dehydrogenase (LDH), lymphocyte count (L), hemoglobin (Hb), D dimer (DD), white blood cells count (WBC), thrombin time (TT) and total protein(TP). Points are assigned for each variable by drawing a line upward from the corresponding variable to the Points line. The sum of points plotted on the total Points line corresponds with the severity of patients with COVID-19.

• 

  Fig. 8. Calibration curves of the radiomics nomogram for predicting the disease progression of COVID-19 pneumonia in validation set (A) and test set (B). The y-axis represents the probability of COVID-19 pneumonia becoming severe, and the x-axis represents the predicted risk. The dashed line was the reference line where an ideal nomogram would lie. The dotted line was the performance of the radiomics nomogram, while the solid line corrected for any bias in the radiomics nomogram. MAE is 0.04 in the validation set and 0.075 in the test set

• 

  Fig. 9. Decision curve analysis (DCA) for the radiomic model, clinical model, and radiomics nomogram. The y-axis measures the net benefit. Using the clinical model, radiomic model, and radiomics nomogram in the study to predict COVID-19 pneumonia progress adds more benefit than the treat all patients as severity patients scheme or the treat none scheme. The net benefit of the radiomics nomogram was better than clinical and radiomic models in both sets, with several overlaps in the training set