Synthesize and Segment: Towards Improved Catheter Segmentation via Adversarial Augmentation

Automatic catheter and guidewire segmentation plays an important role in robot-assisted interventions that are guided by fluoroscopy. Existing learning based methods addressing the task of or tracking often limited scarcity annotated samples difficulty data collection. In case deep methods, demand for large amounts labeled further impedes successful application. We propose a synthesize segment approach with plug possibilities to address this. show adversarially learned image-to-image translation network can catheters X-ray fluoroscopy enabling augmentation order alleviate low regime. To make realistic synthesized images, we train via perceptual loss coupled similarity constraints. Then existing networks used learn accurate localization semi-supervised setting generated images. The empirical results on collected medical datasets value our significant improvements over baseline methods.