Neuroimaging-based Artificial Intelligence (AI) models are being developed to support the early diagnosis of neurological and psychiatric disorders, such as Alzheimer's disease, Parkinson's disease, autism spectrum disorder, and schizophrenia. With powerful inferential capability for complex mappings, deep learning may achieve significant performance improvements over traditional machine learning in neuroimaging analysis. Radiologists benefit from several clinical applications of the research, gaining a better understanding of brain medical images, such as the fusion of fMRI and DTI data for both qualitative and quantitative assessment of diseases from neuroimmune, degenerative, and cerebrovascular diseases. Most computer-aided diagnosis (CAD) models are still in the exploratory stage and have not been validated in real clinical scenarios with ever-changing disease states and complex data environments. This research topic aims to accelerate the translation of these studies on data analysis and models into clinical applications that benefit the diagnosis, treatment, and prognosis of patients. This topic welcomes summaries of the new challenges of AI in neuroimaging data application and encourages discussions on future challenge-solving strategies.
At present, the issues faced by the application of AI in neuroimaging mainly include:
- The black box of deep learning. CAD system designers lack insight into how to make decisions on models, resulting in lack of trust in CAD applications in real world.
- Lack of sufficient data. Neuroimaging is more challenging to acquire than natural images, which makes it hard to collect massive labeled datasets for each disease.
- Data heterogeneity among multiple sites. Heterogeneity can emanate from different scanning equipment and scanning parameters among different sites, resulting in poor generalization ability of AI applications.
- Ethical issues of AI. It is hard to ensure that the constructed AI model will not replicate the stereotypes in the original clinical scene, such as gender discrimination, and racial discrimination.
This Research Topic welcomes theoretical and empirical research in neuroimaging data analysis, with a particular focus on developing and optimizing neuroimaging AI, including but not limited to the following applications:
- Various decision support models in the diagnosis, treatment, and prognosis of neurological and psychiatric disorders.
- Few-shot learning tasks in neuroimaging, including few-shot learning algorithms, transfer learning, and data augmentation.
- More interpretable AI model for neuroimaging.
- AI applications developed in neuroimaging have equal diagnostic performance in different subgroups, e.g., female/male, older/younger.
Neuroimaging-based Artificial Intelligence (AI) models are being developed to support the early diagnosis of neurological and psychiatric disorders, such as Alzheimer's disease, Parkinson's disease, autism spectrum disorder, and schizophrenia. With powerful inferential capability for complex mappings, deep learning may achieve significant performance improvements over traditional machine learning in neuroimaging analysis. Radiologists benefit from several clinical applications of the research, gaining a better understanding of brain medical images, such as the fusion of fMRI and DTI data for both qualitative and quantitative assessment of diseases from neuroimmune, degenerative, and cerebrovascular diseases. Most computer-aided diagnosis (CAD) models are still in the exploratory stage and have not been validated in real clinical scenarios with ever-changing disease states and complex data environments. This research topic aims to accelerate the translation of these studies on data analysis and models into clinical applications that benefit the diagnosis, treatment, and prognosis of patients. This topic welcomes summaries of the new challenges of AI in neuroimaging data application and encourages discussions on future challenge-solving strategies.
At present, the issues faced by the application of AI in neuroimaging mainly include:
- The black box of deep learning. CAD system designers lack insight into how to make decisions on models, resulting in lack of trust in CAD applications in real world.
- Lack of sufficient data. Neuroimaging is more challenging to acquire than natural images, which makes it hard to collect massive labeled datasets for each disease.
- Data heterogeneity among multiple sites. Heterogeneity can emanate from different scanning equipment and scanning parameters among different sites, resulting in poor generalization ability of AI applications.
- Ethical issues of AI. It is hard to ensure that the constructed AI model will not replicate the stereotypes in the original clinical scene, such as gender discrimination, and racial discrimination.
This Research Topic welcomes theoretical and empirical research in neuroimaging data analysis, with a particular focus on developing and optimizing neuroimaging AI, including but not limited to the following applications:
- Various decision support models in the diagnosis, treatment, and prognosis of neurological and psychiatric disorders.
- Few-shot learning tasks in neuroimaging, including few-shot learning algorithms, transfer learning, and data augmentation.
- More interpretable AI model for neuroimaging.
- AI applications developed in neuroimaging have equal diagnostic performance in different subgroups, e.g., female/male, older/younger.