FusGAN: GAN-Based Ultrasound Simulation from CT Slices

FusGAN: GAN-Based Ultrasound Simulation from CT Slices - Complete Dataset and Models Description This repository contains the complete implementation, trained models, and datasets for FusGAN, a deep learning system that generates realistic ultrasound simulations from CT slices using Generative Adversarial Networks (GANs). The system takes CT scan data and transducer mask inputs to produce high-quality ultrasound intensity maps, enabling medical imaging research and training applications. Contents This Zenodo package includes: Source Code: Complete Python implementation with training and inference scripts Pre-trained Models: Trained GAN models ready for ultrasound simulation generation Training Dataset: MATLAB (.mat) files containing: CT slice arrays (ct_slices) Transducer mask arrays (transducer_masks) Ground truth ultrasound simulation arrays (pi_maps) Example Results: Sample outputs demonstrating the simulation quality Technical Specifications Framework: TensorFlow/Keras Input Format: MATLAB (.mat) files containing arrays for CT slices, transducer masks, and ultrasound maps Output Format: Simulated ultrasound intensity maps Model Architecture: Generative Adversarial Network optimized for medical imaging Usage The package provides both training capabilities for custom datasets and pre-trained models for immediate use. Users can either train new models with their own data or utilize the provided trained models for ultrasound simulation generation. License MIT License - see LICENSE file for details. Contact Agustin Conesa ([e-mail]) Keywords ultrasound simulation, medical imaging, generative adversarial networks, CT imaging, deep learning, medical AI, image synthesis, transducer modeling