2018/03/25 by Harrison Nguyen, Nguyen, Harrison, Richard W. Morris +8
Computer Science · Medicine · #Advanced Neuroimaging Techniques and Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Radiomics and Machine Learning in Medical Imaging
paper · pdf · doi:10.48550/arxiv.1803.09375
openalex publication_date 2018/03/25 · openalex created_date 2022/09/14 · openalex updated_date 2026/07/28
Magnetic Resonance Imaging (MRI) of the brain has been used to investigate a\nwide range of neurological disorders, but data acquisition can be expensive,\ntime-consuming, and inconvenient. Multi-site studies present a valuable\nopportunity to advance research by pooling data in order to increase\nsensitivity and statistical power. However images derived from MRI are\nsusceptible to both obvious and non-obvious differences between sites which can\nintroduce bias and subject variance, and so reduce statistical power. To\nrectify these differences, we propose a data driven approach using a deep\nlearning architecture known as generative adversarial networks (GANs). GANs\nlearn to estimate two distributions, and can then be used to transform examples\nfrom one distribution into the other distribution. Here we transform\nT1-weighted brain images collected from two different sites into MR images from\nthe same site. We evaluate whether our model can reduce site-specific\ndifferences without loss of information related to gender (male, female) or\nclinical diagnosis (schizophrenia, bipolar disorder, healthy). When trained\nappropriately, our model is able to normalise imaging sets to a common scanner\nset with less information loss compared to current approaches. An important\nadvantage is our method can be treated as a black box that does not require any\nknowledge of the sources of bias but only needs at least two distinct imaging\nsets.\n