Related Experiment Video
Updated: Sep 7, 2025
08:49
Whole-Brain Single-Cell Imaging and Analysis of Intact Neonatal Mouse Brains Using MRI, Tissue Clearing, and Light-Sheet Microscopy
Published on: August 1, 2022
3.7K
Deep Learning in Neuroimaging: Overcoming Challenges With Emerging Approaches.
Jason Smucny1, Ge Shi2, Ian Davidson2
1Department of Psychiatry and Behavioral Sciences, University of California, Davis, Davis, CA, United States.
Frontiers in Psychiatry
|June 20, 2022
Summary
Three novel deep learning (DL) methods, including transfer learning, data augmentation, and explainable AI (XAI), can improve psychiatric research. These techniques address DL
Area of Science:
- Psychiatric research
- Neuroimaging
- Artificial intelligence
Background:
- Deep learning (DL) shows promise for psychiatric research by analyzing complex datasets like fMRI to predict clinical outcomes.
- Traditional DL methods face challenges in medical imaging due to requirements for large datasets and model opaqueness, hindering clinical application.
Purpose of the Study:
- To introduce novel DL approaches that facilitate the integration of DL into psychiatric research and clinical practice as prognostic tools.
- To address limitations of current DL methods, specifically the need for extensive data and the 'black box' problem.
Main Methods:
- Transfer learning: Adapting knowledge from one dataset (e.g., fMRI data from one site) to another (e.g., data from a different site).
- Data augmentation (via Mixup): Creating 'virtual' data instances to increase dataset size and improve model accuracy.
- Explainable Artificial Intelligence (XAI): Utilizing tools to interpret DL model decisions, revealing key features and their combinations.
Main Results:
- Transfer learning and Mixup-based data augmentation reduce the amount of training data needed for accurate DL models in psychiatric research.
- XAI provides insights into DL decision-making processes, addressing the 'black box' issue and revealing underlying clinical outcome mechanisms.
Conclusions:
- These DL techniques enhance the applicability of AI in psychiatry, potentially accelerating diagnostic and prognostic tool development.
- The methods discussed can uncover novel biological and clinical mechanisms underlying mental illness, paving the way for new therapeutic interventions.
