Synthesising Clinically Realistic Chest X-Rays using Generative Adversarial Networks
Chest x-rays are a vital diagnostic tool in the workup of many patients. Similar to most medical imaging modalities, they are profoundly multi-modal and are capable of visualising a variety of combinations of conditions. There is an ever pressing need for greater quantities of labelled images to drive forward the development of diagnostic tools, however this is in direct opposition to concerns regarding patient confidentiality which constrains access through permission requests and ethics approvals. Previous work has sought to address these concerns by creating class-specific generative adversarial networks (GANs) that synthesise images to augment training data. These approaches cannot be scaled as they introduce computational trade offs between model size and class number which places fixed limits on the quality that such generates can achieve. This project addresses these concerns by introducing latent class optimisation which enables efficient, multi-modal sampling from a GAN and with which a large archive of labelled generates is synthesised. This project applies a Progressive Growing GAN (PGGAN) to the task of unsupervised x-ray synthesis and has radiologists evaluate the clinical realism of the resultant samples. An in depth review of the properties of varying pathologies seen on generates as well as an overview of the extent of disease diversity captured by the model is provided. The application of the Fréchet Inception Distance (FID) to measure the quality of x-ray generates and find that they are similar to other high resolution tasks is also validated. X-ray clinical realism is quantified by asking radiologists to distinguish between real and synthesised scans with the finding that generates are more likely to be classed as real than by chance, however, there remains progress required to achieve true realism. These findings are confirmed through the evaluation of a synthetic classification model performance on real scans. The project concludes with a discussion the limitations of PGGAN generates and how to achieve controllable, realistic generates in future work.
A reaserch report submitted in partial fulfilment of the requirements for the degree of Master of Science in Engineering to the Faculty of Engineering and the BuiltEnvironment, School of Electrical and Information Engineering, University of the Witwatersrand, Johannesburg, 2021