Fig 1.
Custom 3D CNN architecture used for the classification model approach.
C: 3x3x3 Conv + LeakyReLU + MaxPooling + Batch Normalization; FC1: 128 neuron fully connected layer; FC2: 3 neuron fully connected layer. CNN, convolutional neuronal network.
Fig 2.
COVID-19, coronavirus disease 2019; CT, computed tomography.
Table 1.
Patient demographics.
Table 2.
Distribution of diagnostic classes for both COVID-19 classification and segmentation models.
Fig 3.
Receiver operating characteristic (ROC) curve for the COVID-19 classification model.
AUC, area under the ROC curve.
Fig 4.
Confusion matrix for the COVID-19 classification model.
COVID-19, coronavirus disease 2019.
Table 3.
Performance metrics for the COVID-19 classification model (COVID-19 vs other cases).
Table 4.
Segmentation model performance metrics.
Fig 5.
Structured report with analysis results.
A 3D reconstruction of the lungs is generated, together with the most affected transverse CT slice and segmentation masks of the lung opacities. The report includes disease probabilities and quantitative analysis results. COVID-19, coronavirus disease 2019; CT, computed tomography.