Table 2 Results summary of all experiments where we train the models on either batch and test the models within the same-batch or cross-batch. We perform hypothesis testing where the null hypothesis states that the model testing accuracy with color normalization is the same as using just the original H&E images and the alternative hypothesis as the model testing performance with color normalization is better than with the original H &E images. p-values are indicated in the parenthesis.

From: Impact of stain variation and color normalization for prognostic predictions in pathology