skip to main content
Idioma:
Primo Search
Clear Search Box
Search in: Busca Geral
Busca Avançada
Busca por Índices

Differentiable neural architecture search for optimal spatial/temporal brain function network decomposition

Li, Qing ; Wu, Xia ; Liu, Tianming

Medical image analysis, 2021-04, Vol.69, p.101974-101974, Article 101974 [Periódico revisado por pares]

Netherlands: Elsevier B.V

Texto completo disponível

Citações Citado por
  • Título:
    Differentiable neural architecture search for optimal spatial/temporal brain function network decomposition
  • Autor: Li, Qing ; Wu, Xia ; Liu, Tianming
  • Assuntos: Artificial neural networks ; Brain ; Brain mapping ; Computer architecture ; Cytology ; Decomposition ; Differentiable neural architecture search ; Functional magnetic resonance imaging ; Magnetic resonance imaging ; Neural networks ; Neuroimaging ; Optimization ; Recurrent neural networks ; Search algorithms ; Search methods ; Spatial/temporal ; Task-based fMRI ; Vanilla
  • É parte de: Medical image analysis, 2021-04, Vol.69, p.101974-101974, Article 101974
  • Notas: ObjectType-Article-1
    SourceType-Scholarly Journals-1
    ObjectType-Feature-2
    content type line 23
  • Descrição: •We propose a novel NAS framework (ST-DARTS) that is suitable for spatial and temporal brain network learning when searching in the continuous space for the cell structure.•The flexible early-stopping mechanism is introduced into our ST-DARTS framework (ST-DARTS+), which can significantly increase the brain network decomposition performance, and address the collapse issue of vanilla DARTS and ST-DARTS caused by over-fitting with asimple but effective way.•Promising and stable results are achieved for 4D fMRI data modeling and brain network decomposition, which proves the effectiveness, efficiency and robustness of our ST-DARTS and ST-DARTS+ models. It has been a key topic to decompose the brain's spatial/temporal function networks from 4D functional magnetic resonance imaging (fMRI) data. With the advantages of robust and meaningful brain pattern extraction, deep neural networks have been shown to be more powerful and flexible in fMRI data modeling than other traditional methods. However, the challenge of designing neural network architecture for high-dimensional and complex fMRI data has also been realized recently. In this paper, we propose a new spatial/temporal differentiable neural architecture search algorithm (ST-DARTS) for optimal brain network decomposition. The core idea of ST-DARTS is to optimize the inner cell structure of the vanilla recurrent neural network (RNN) in order to effectively decompose spatial/temporal brain function networks from fMRI data. Based on the evaluations on all seven fMRI tasks in human connectome project (HCP) dataset, the ST-DARTS model is shown to perform promisingly, both spatially (i.e., it can recognize the most stimuli-correlated spatial brain network activation that is very similar to the benchmark) and temporally (i.e., its temporal activity is highly positively correlated with the task-design). To further improve the efficiency of ST-DARTS model, we introduce a flexible early-stopping mechanism, named as ST-DARTS+, which further improves experimental results significantly. To our best knowledge, the proposed ST-DARTS and ST-DARTS+ models are among the early efforts in optimally decomposing spatial/temporal brain function networks from fMRI data with neural architecture search strategy and they demonstrate great promise. Framework of spatial/temporal differentiable neural architecture search algorithm (ST-DARTS) ((a) - (d)) and ST-DARTS+ ((a) - (e)) models. (a) the spatial-temporal fMRI data inputs that consist of the spatial (360 ROIs (region of interest)) and temporal (length of fMRI tasks’ time points) dimensions; (b) the predicted spatial-temporal fMRI outputs that will be evaluated by the cross-entropy loss; (c) the first step of the two-stage training strategy: cell structure learning; (d) the second step of the two-stage training strategy: decomposed spatial/temporal function networks learning. The evaluation processes are the same for both stages. [Display omitted]
  • Editor: Netherlands: Elsevier B.V
  • Idioma: Inglês
Links
Voltar para lista de resultados

  • Realização: ABCD-USP USP   Logos de Redes Sociais: Freepik

Buscando em bases de dados remotas. Favor aguardar.