A Review of Generative Adversarial Networks in Cancer Imaging: New Applications, New Solutions

Abstract

Despite technological and medical advances, the detection, interpretation, and treatment of cancer based on imaging data continue to pose significant challenges. These include high inter-observer variability, difficulty of small-sized lesion detection, nodule interpretation and malignancy determination, inter-and intra-tumour heterogeneity, class imbalance, segmentation inaccuracies, and treatment effect uncertainty. The recent advancements in Generative Adversarial Networks (GANs) in computer vision as well as in medical imaging may provide a basis for enhanced capabilities in cancer detection and analysis. In this review, we assess the potential of GANs to address a number of key challenges of cancer imaging, including data scarcity and imbalance, domain and dataset shifts, data access and privacy, data annotation and quantification, as well as cancer detection, tumour profiling and treatment planning. We provide a critical appraisal of the existing literature of GANs applied to cancer imagery, together with suggestions on future research directions to address these challenges. We analyse and discuss 163 papers that apply adversarial training techniques in the context of cancer imaging and elaborate their methodologies, advantages and limitations. With this work, we strive to bridge the gap between the needs of the clinical cancer imaging community and the current and prospective research on GANs in the artificial intelligence community.

FUTURE-AI: Guiding Principles and Consensus Recommendations for Trustworthy Artificial Intelligence in Medical Imaging

“…39. Image harmonisation: If differences in imaging and acquisition protocols cannot be prevented between centres, robustness should be enhanced by implementing image harmonisation tools and techniques such as histogram normalisation and discretisation [112], ComBat harmonisation [45,139], and data augmentation solutions with neural style transfer methods, Generative Adversarial Networks and unsupervised image-to-image translation units [47,171,122]. It is recommended to assess and report the variation across features alongside the reduction in variation after applying harmonisation methods to these features in the dataset.…”

Sharing Generative Models Instead of Private Data: A Simulation Study on Mammography Patch Classification

“…However, MMG images are limited by their error rates due to tissue superposition which could lead to underdiagnosis of significant breast cancers (false negatives) and overdiagnosis of insignificantly abnormal or healthy cases (false positives) [1,23]. In this regard, deep learning based computer-aided detection (CADe) systems have shown great promise in improving and automating the decision making process of mammograms [6,1].…”

“…A solution to this problem is inter-centre data sharing. However, clinical centres are constrained from sharing sensitive patient data due to technical, legal, and most importantly, ethical concerns [7,23].…”

“…Synthetic images generated by Generative Adversarial Networks (GANs) [13] have been shown to be of high perceived visual realism for mammography and to improve downstream tasks including cancer detection, tumour segmentation and classification [23]. Therefore, in this work, we hypothesize that GANs can overcome the inter-centre data sharing constraints.…”

“…Therefore, in this work, we hypothesize that GANs can overcome the inter-centre data sharing constraints. After learning the real data distribution, GANs can generate synthetic data with limited risk of revealing sensitive patient information [27,23]. Then, a clinical centre can share a trained generative model that will act as a proxy for the real patient data.…”