Reconstructing lost BOLD signal in individual participants using deep machine learning

• DOI: 10.1038/s41467-020-18823-9
• Publication Date: 2020-10-07T10:02:46Z
• Authors: Yuxiang Yan; Louisa Dahmani; Jianxun Ren; Lunhao Shen; Xiaolong Peng; Ruiqi Wang; Changgeng He; Changqing Jiang; Chen Gong; Ye Tian; Jianguo Zhang; Yi Guo; Yuanxiang Lin; Shijun Li; Meiyun Wang; Luming Li; Bo Hong; Hesheng Liu
• Keywords: SIGNAL (programming language); Blood-oxygen-level dependent

Abstract Signal loss in blood oxygen level-dependent (BOLD) functional neuroimaging is common and can lead to misinterpretation of findings. Here, we reconstructed compromised fMRI signal using deep machine learning. We trained a model learn principles governing BOLD activity one dataset reconstruct artificially regions an independent dataset, frame by frame. Intriguingly, time series extracted from frames are correlated with the original series, even though do not independently carry any temporal information. Moreover, connectivity maps exhibit good correspondence maps, indicating that recovers relationships among brain regions. replicated this result two healthy datasets patients whose scans suffered due intracortical electrodes. Critically, reconstructions capture individual-specific Deep learning thus presents unique opportunity while capturing features individual’s own organization.