Simulating Patho-realistic Ultrasound Images using Deep Generative Networks with Adversarial Learning

摘要

Ultrasound imaging makes use of backscattering of waves during theirinteraction with scatterers present in biological tissues. Simulation ofsynthetic ultrasound images is a challenging problem on account of inability toaccurately model various factors of which some include intra-/inter scanlineinterference, transducer to surface coupling, artifacts on transducer elements,inhomogeneous shadowing and nonlinear attenuation. Current approaches typicallysolve wave space equations making them computationally expensive and slow tooperate. We propose a generative adversarial network (GAN) inspired approachfor fast simulation of patho-realistic ultrasound images. We apply theframework to intravascular ultrasound (IVUS) simulation. A stage 0 simulationperformed using pseudo B-mode ultrasound image simulator yields speckle mappingof a digitally defined phantom. The stage I GAN subsequently refines them topreserve tissue specific speckle intensities. The stage II GAN further refinesthem to generate high resolution images with patho-realistic speckle profiles.We evaluate patho-realism of simulated images with a visual Turing testindicating an equivocal confusion in discriminating simulated from real. Wealso quantify the shift in tissue specific intensity distributions of the realand simulated images to prove their similarity.