eduzhai > Applied Sciences > Engineering >

Deep Image Translation for Enhancing Simulated Ultrasound Images

  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Ultrasound simulation based on ray tracing enables the synthesis of highlyrealistic images. It can provide an interactive environment for trainingsonographers as an educational tool. However, due to high computational demand,there is a trade-off between image quality and interactivity, potentiallyleading to sub-optimal results at interactive rates. In this work we introducea deep learning approach based on adversarial training that mitigates thistrade-off by improving the quality of simulated images with constantcomputation time. An image-to-image translation framework is utilized totranslate low quality images into high quality versions. To incorporateanatomical information potentially lost in low quality images, we additionallyprovide segmentation maps to image translation. Furthermore, we propose toleverage information from acoustic attenuation maps to better preserve acousticshadows and directional artifacts, an invaluable feature for ultrasound imageinterpretation. The proposed method yields an improvement of 7.2 inFréchet Inception Distance and 8.9 in patch-based Kullback-Leiblerdivergence.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...