Adversarial Data Augmentation on Breast MRI Segmentation

The scarcity of balanced and annotated datasets has been a recurring problem in medical image analysis. Several researchers have tried to fill this gap employing dataset synthesis with adversarial networks (GANs). Breast magnetic resonance imaging (MRI) provides complex, texture-rich images, the same annotation shortage issues, for which, best our knowledge, no previous work synthesizing data. Within context, addresses breast MRI images from corresponding annotations evaluate impact data augmentation strategy on semantic segmentation task. We explored variations image-to-image translation using conditional GANs, namely fitting generator’s architecture residual blocks experimenting cycle consistency approaches. studied these changes visual verisimilarity how an U-Net model is affected by usage synthetic achieved sufficiently realistic-looking maintained stable score even when completely replacing set. Our results were promising, especially concerning Pix2PixHD Residual CycleGAN architectures.