loading page

Multimodal Cross Enhanced Fusion Network with Multiscale Long-range Reception for Diagnosis of Subjective Memory Complaints
  • +6
  • Yilin Leng ,
  • Wenjun Cui ,
  • Yunsong Peng ,
  • Caiying Yan ,
  • Yuzhu Cao ,
  • Zhuangzhi Yan ,
  • Shuangqing Chen ,
  • Xi Jiang ,
  • Jian Zheng
Yilin Leng
Shanghai University, Shanghai University

Corresponding Author:[email protected]

Author Profile
Wenjun Cui
Author Profile
Yunsong Peng
Author Profile
Caiying Yan
Author Profile
Yuzhu Cao
Author Profile
Zhuangzhi Yan
Author Profile
Shuangqing Chen
Author Profile
Jian Zheng
Author Profile


Many deep learning methods such as convolutional neural networks have been proposed for multimodal diagnosis of AD and its early stages (SMC, subjective memory complaints), which may help to intervene in the treatment to slow its progression. However, it is challenging to obtain long-range dependencies of the whole brain by convolutional networks due to the limited receptive field, and most methods introduce patch embedding and attention to obtain global information which require large computational effort and perform poorly on small datasets. Moreover, multimodal fusion methods are generally coarse, and their use of shared extractors or simple downscaling stitching may lead to suboptimal results. In this work, we propose a novel network called MENet, which adaptively recalibrates multiscale long-range receptive field guided by channel attention to localize discriminative brain regions. Based on this, the network extracts the response between sMRI and FDG-PET as an enhancement for AD and SMC diagnosis. Our proposed method was evaluated on the publicly released ADNI datasets, and achieves state-of-the-art performance in both diagnosis of AD and SMC with sMRI and FDG-PET. To the best of our knowledge, this is one of the pioneering studies to classify SMC vs. NC with FDG-PET based on a three-dimensional deep learning approach.
May 2023Published in Computers in Biology and Medicine volume 157 on pages 106788. 10.1016/j.compbiomed.2023.106788