November 2023
·
39 Reads
Communications in Computer and Information Science
Radiogenomics is a newly emerging field that integrates genomics with medical imaging data. The aim of this field is to elucidate the associations between gene expression data and imaging phenotypes, especially in cancer. However, radiogenomics is hindered by the expensive cost of genetic screening tests which leads to the unavailability of numerous big datasets for paired imaging and genetic data. Big data is crucial for training machine learning-based techniques for analysis relating to radiogenic studies. Currently, fake data is generated on only one of the two radiogenomic types; genomic or medical imaging data are generated separately. In this paper, we propose a deepfake approach implemented by combining two Generative Adversarial Networks (GANs) to create fake image data that are hardly differentiable from the original to improve Breast cancer diagnosis. To evaluate the model, a survey is developed and distributed among the participants to measure their ability to differentiate the original from the Deepfakes images. The results showed that the model-generated fake images cannot be distinguished from the authentic images and are relatively satisfying using the PyTorch framework.