Deep learning for brain disorders

By Ninon Burgos; Simona Bottani; Johann Faouzi; Elina Thibeau-Sutre; Olivier Colliot

Keywords: Deep learning Neurology Medical imaging Genomics

This review will enable readers to grasp the full potential of deep learning for brain disorders as it presents the main uses of deep learning all along the medical data analysis chain: from data acquisition to disease treatment. We first focus on data processing, covering image reconstruction, signal enhancement and cross-modality image synthesis, and on the biomarkers that can be extracted from spatio-temporal neuroimaging data, such as the volume of normal structures or of lesions. We then describe how deep learning can be used to detect diseases, predict their evolution, improve their understanding and help develop treatments. For these applications, we emphasize the types of architectures and data used, as well as the concerned disorders. Finally, we highlight trending applications and provide guidelines to bridge the gap between research studies and clinical routine.

Published in Briefings in Bioinformatics the 15/12/2020

Common deep learning architectures for brain disorders. a) U-Net is the most popular architecture for biomedical image segmentation. U-Net architectures have also been used for image reconstruction and synthesis. b) Autoencoders have been used for disease detection, prediction of treatment and integration of multimodal data. c) Variational autoencoders have been used for image segmentation, disease detection and disease subtyping. d) Generative adversarial networks can be used for data augmentation. e) Conditional generative adversarial networks have been used for signal enhancement, image synthesis and disease prediction.

Full title Deep learning for brain disorders: from data processing to disease treatment

Article abstract:

In order to reach precision medicine and improve patients’ quality of life, machine learning is increasingly used in medicine. Brain disorders are often complex and heterogeneous, and several modalities such as demographic, clinical, imaging, genetics and environmental data have been studied to improve their understanding. Deep learning, a subpart of machine learning, provides complex algorithms that can learn from such various data. It has become state-of-the-art in numerous fields including computer vision and natural language processing, and is also growingly applied in medicine. In this article, we review the use of deep learning for brain disorders. More specifically, we identify the main applications, the concerned disorders and the types of architectures and data used. Finally, we provide guidelines to bridge the gap between research studies and clinical routine.

Summary key points

  • Deep learning has been applied to various tasks related to brain disorders, such as image reconstruction, synthesis and segmentation, or disease diagnosis and outcome prediction.
  • Convolutional neural networks have been successfully applied to imaging and genetic data in numerous brain disorders, while recurrent neural networks showed encouraging results with longitudinal clinical data and sensor data.
  • Despite the promising results obtained with deep learning, several important limitations need addressing before an application in clinical routine becomes possible.
  • Future research should especially focus on the generalizability and interpretability of deep learning models.