Classification performance bias between training and test sets in a limited mammography dataset.

Performance bias can result from model testing when using limited datasets. Optimal strategies for test set selection should be developed to ensure study conclusions are appropriate.