Over the past decades, neuroimaging techniques have revolutionized our understanding of brain structure, function, organization, as well as neurological and psychiatric disorders. Machine learning has emerged as a crucial tool in diverse neuroimaging applications, such as image segmentation and registration, anatomical measurements, multivariate pattern analysis, signal decomposition, disease diagnosis, biomarker identification, etc.
The last decade witnessed a rapid advancement of deep learning in various aspects such as learning theories, neural network architecture, and deep learning infrastructure. Furthermore, open data initiatives such as the Human Connectome Project (HCP) have made large amounts of data available to researchers. These efforts open up new opportunities in neuroimaging data analysis.
However, we are faced with several challenges when applying deep learning techniques to neuroimaging tasks. One challenge is the lack of and sometimes unavailability of labels. For example, in image and signal denoising, clean ground-truth data are not available. Another challenge is imbalanced sample size: patients usually consist of a small portion of the population and open datasets tend to contain more healthy subjects than patients. Unsupervised deep learning is an important tool to solve these challenges. Furthermore, a number of traditional topics in neuroimaging such as brain parcellation and signal decomposition fall under the umbrella of unsupervised machine learning. It is therefore meaningful to revisit them with the state-of-the-art unsupervised deep learning techniques.
This Research Topic calls for original research contributions to unsupervised deep learning for neuroimaging data and reviews of this topic, including but not limited to the following themes:
Applying unsupervised deep learning to neuroimaging data processing such as image registration and segmentation, image and signal denoising, etc.
Applying unsupervised deep learning to reveal brain organizations and networks such as brain parcellation, disentangled representation learning, manifold learning, pattern mining, etc.
How can the representations and patterns derived from neuroimaging data be translated into meaningful neuroscience knowledge.
Transfer learning to mitigate challenges posed by limited labeled data. For example, learning representations on unlabeled public datasets and applying them to smaller datasets with labels.
Keywords:
unsupervised deep learning, neuroimaging, explainable AI, transfer learning, neuroimaging data processing, brain networks, brain organization, data mining in neuroimaging
Important Note:
All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.
Over the past decades, neuroimaging techniques have revolutionized our understanding of brain structure, function, organization, as well as neurological and psychiatric disorders. Machine learning has emerged as a crucial tool in diverse neuroimaging applications, such as image segmentation and registration, anatomical measurements, multivariate pattern analysis, signal decomposition, disease diagnosis, biomarker identification, etc.
The last decade witnessed a rapid advancement of deep learning in various aspects such as learning theories, neural network architecture, and deep learning infrastructure. Furthermore, open data initiatives such as the Human Connectome Project (HCP) have made large amounts of data available to researchers. These efforts open up new opportunities in neuroimaging data analysis.
However, we are faced with several challenges when applying deep learning techniques to neuroimaging tasks. One challenge is the lack of and sometimes unavailability of labels. For example, in image and signal denoising, clean ground-truth data are not available. Another challenge is imbalanced sample size: patients usually consist of a small portion of the population and open datasets tend to contain more healthy subjects than patients. Unsupervised deep learning is an important tool to solve these challenges. Furthermore, a number of traditional topics in neuroimaging such as brain parcellation and signal decomposition fall under the umbrella of unsupervised machine learning. It is therefore meaningful to revisit them with the state-of-the-art unsupervised deep learning techniques.
This Research Topic calls for original research contributions to unsupervised deep learning for neuroimaging data and reviews of this topic, including but not limited to the following themes:
Applying unsupervised deep learning to neuroimaging data processing such as image registration and segmentation, image and signal denoising, etc.
Applying unsupervised deep learning to reveal brain organizations and networks such as brain parcellation, disentangled representation learning, manifold learning, pattern mining, etc.
How can the representations and patterns derived from neuroimaging data be translated into meaningful neuroscience knowledge.
Transfer learning to mitigate challenges posed by limited labeled data. For example, learning representations on unlabeled public datasets and applying them to smaller datasets with labels.
Keywords:
unsupervised deep learning, neuroimaging, explainable AI, transfer learning, neuroimaging data processing, brain networks, brain organization, data mining in neuroimaging
Important Note:
All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.