Deep Learning for Pathology is known to be susceptible to robustness issues. Robustness refers to how reliably and accurately deep learning models behave or perform across different types of data; e.g. different scanner vendors, hospitals, or patient groups. In this study we aim to:

1. Define proper metrics with which robustness can be measured.
2. Consider scenarios which may affect model performance such as:

• Sites (labs)
• Staining procedure
• Scanner and sensor type
• Rotations

3. Improve model robustness for WSI classification and segmentation tasks with a novel method.For this project, I collaborate with Kaiko.ai.