What intensity range should we use for the lung images? Should we include attention layers in the model?