Reply to Mummadi et al.: Overfitting and Use of Mismatched Cohorts in Deep Learning Models: Preventable Design Limitations