Strategically selecting a combination of synthetic tumor sizes and generating synthetic tumors with precise boundaries significantly improves the accuracy of deep learning-based pancreatic tumor segmentation models.
