AUC/F1 as score metric in simple convnet for classification on mnist dataset

Question

I am doing classification on MNIST dataset using very shallow convnet architecture but instead of accuracy as metric to compare train and validation score at each epoch, i wish