This was my graduation project, on supporting COVID-19 diagnosis from lung MRI images. To be clear about the scope: it's a study project, not a clinical tool.
the model is a convolutional neural network built with TensorFlow and Keras, classifying images across several diagnostic categories rather than a simple yes or no. Before training, the images went through preprocessing and scaling, which matters because a CNN is sensitive to how its input is normalized.
tuning was mostly about architecture: pooling layers to reduce the feature maps, and dropout layers to reduce overfitting.
I evaluated the trained model on the training, validation and test sets separately, looking at categorical cross-entropy loss and accuracy. Comparing those three is what tells you whether the model learned the task or just memorized the training data. the final model reached 94.5% accuracy on the test set.