Identification of glioblastoma molecular subtype and prognosis based on deep MRI features
作者：
Highlights：
• We divided the radiogenomics study into imaging analysis stage and bio-statistical learning stage and gave a complete review on both stages respectively.
• We developed a predictive model, named DeepRA, based on deep imaging features to identify MRI signatures for accurate prediction of GBM molecular subtype and patient overall survival.
• We converted the high-dimensional feature representations extracted from deep networks to a flat feature vector. The feature vector is more interpretable and efficient.
• We validated the DeepRA on TCGA data. Comparisons with hand-crafted features and other deep feature show the effectiveness of DeepRA.
摘要
•We divided the radiogenomics study into imaging analysis stage and bio-statistical learning stage and gave a complete review on both stages respectively.•We developed a predictive model, named DeepRA, based on deep imaging features to identify MRI signatures for accurate prediction of GBM molecular subtype and patient overall survival.•We converted the high-dimensional feature representations extracted from deep networks to a flat feature vector. The feature vector is more interpretable and efficient.•We validated the DeepRA on TCGA data. Comparisons with hand-crafted features and other deep feature show the effectiveness of DeepRA.
论文关键词：Deep MRI features,Glioblastoma multiforme,Molecular subtype,Survival,Feature representation
论文评审过程：Received 2 June 2020, Revised 8 September 2021, Accepted 10 September 2021, Available online 16 September 2021, Version of Record 28 September 2021.
论文官网地址：https://doi.org/10.1016/j.knosys.2021.107490
