Algorithm 1: BNCNN model

INPUT IMG: Dataset of X-ray images [224 × 224 × 15,153] LAB: Set of labels {‘COVID-19’, ‘Normal’, ‘Viral Pneumonia’} corresponding to X-ray images [1 × 15,153] $N_{class}$: number of classes used for training OUTPUT $Y_{pred}$: a matrix of prediction probability of class labels [$N_{class}$ × 683] ALGORITHM: If 3-way $N_{class}$ = 3 If 2-way $N_{class}$ = 2 For I = 1 to length (LAB) If LAB (1, i) == ‘COVID-19’ or LAB(1, i) == ‘Normal’ X(:, :, i), Y(1, i) = IMG(:, :, i), LAB(1, i) End if End_for Else $N_{class}$ = 2 X, Y = IMG, LAB End_if X = image_resize(X)/255 $Y_{enc}$ = one-hot_encoder (Y, $N_{class}$) If model_name == ‘VGG-16’ or model_name == ‘VGG-16’ model = load_pre-trained_weigts(model_name) model = freeze_feature_extration_layers(model) Else_if model_name == ‘BNCNN’ model = construct_model() End_if $X_{train}$, $X_{val}$, $X_{test}$, $Y_{train}$, $Y_{val}$, $Y_{test}$ = cross-validation (X, $Y_{enc}$, train = 0.8, val = 0.1, test = 0.1) $X_{train\_aug}$, $Y_{train\_aug}$ = data_augmentation($X_{train}$, $Y_{train}$, rotation = [−20, 20], zoom = [0, 0.2], shear = [0, 0.2], horizontal_flip = True, vertical_flip = False) $X_{val\_aug}$, $Y_{val\_aug}$ = data_augmentation($X_{train}$, $Y_{train}$, rotation = [−20, 20], zoom = [0, 0.2], Shear = [0, 0.2], horizontal_flip = True, vertical_flip = False) model = train ($X_{train\_aug}$, $Y_{train\_aug}$, $X_{val\_aug}$, $Y_{val\_aug}$, custom_callbacks) $Y_{pred}$ = predict (model, $X_{test}$) cm = calculate_confusion_matrix($Y_{pred}$, $Y_{test}$, $N_{class}$) acc, recall, precision, F1-score = calculate_evaluation_metrics(cm, $N_{class}$)