EP01.51: Artificial intelligence in the optimisation of the quality of ultrasound images

To describe the method for optimising the quality of cardiac ultrasound images (UltraGAN). Descriptive study on obstetric ultrasound images at a University Hospital from 2019 to 2022. In this first phase, a public database of heart ultrasound images called CAMUS (Leclerc, 2019) was used. The CAMUS database contains ultrasound images of 450 patients, of which 309 are at pathological risk. Considering that image quality and the expertise of the ultrasonographer are closely related to the diagnosis, the UltraGAN method was used. This method is based on an artificial intelligence tool to improve the quality of the images. The method involves using high-quality image features to train a convolutional neural network (CNN) and transfer this information to medium and low-quality images. For the development of this methodology, generative adversarial networks (GANs) were used (Goodfellow, 2014), which considered the anatomical characteristics of low-quality image regions to maintain spatial consistency (Isola, 2017). The optimisation of cardiac images through artificial intelligence (UltraGAN) allows for the automatic identification of structures present in ultrasounds with an accuracy of over 90% without compromising the anatomical characteristics of the original images, demonstrating its potential as a complementary tool in the screening of congenital heart diseases. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.