• G8_332117
  • 26.5MB
  • zip
  • 0
  • VIP专享
  • 0
  • 2022-05-16 08:30
matlab精度检验代码Python3中的BGSR(脑图超分辨率) BGSR(脑图超分辨率),用于超分辨率的低分辨率图代码,由Busra Asan重新编码。 请联系进一步查询。 谢谢。 You can download the Matlab version of BGSR at: 尽管已经提出了一些用于MRI超分辨率的图像超分辨率技术,但是目前还没有图超分辨率技术。 为此,我们使用功能性大脑数据设计了第一个脑图超分辨率,旨在促进神经系统疾病的诊断。 我们的框架学习如何从低分辨率(LR)图生成高分辨率(HR)图,而无需诉诸用于高分辨率规模的Connectome构建的计算昂贵的图像处理管道。 拟议的详细BGSR管道 这项工作已在杂志上发表。 脑图超分辨率(BGSR)是第一个脑图超分辨率。 所提出的方法包括三个步骤,即通过非线性融合训练样本中的HR图来构建测试对象的HR图。(1)第一步是多核流形学习,以对相似的低分辨率(LR)进行聚类)中使用一组高斯核的权重绘制训练样本中的图表。 (2)第二步是使用带有核矩阵P的迭代扩散过程
# BGSR (Brain Graph Super-Resolution) in Python3 BGSR (Brain Graph Super-Resolution) for super-resolving low-resolution graphs code, recoded by Busra Asan. Please contact for further inquiries. Thanks. ``` You can download the Matlab version of BGSR at: ``` While a few image super-resolution techniques have been proposed for MRI super-resolution, graph super-resolution techniques are currently absent. To this aim, we design the first brain graph super-resolution using functional brain data with the aim to boost neurological disorder diagnosis. Our framework learns how to generate high-resolution (HR) graphs from low-resolution (LR) graphs without resorting to the computationally expensive image processing pipelines for connectome construction at high-resolution scales. ![BGSR pipeline]( # Detailed proposed BGSR pipeline This work has been published in the journal. Brain Graph Super Resolution (BGSR) is the first brain graph super resolution. The proposed method consists of three steps to build a HR graph of a testing subject by non-linearly fusing the HR graphs in the training sample.(1) The first step is a multi-kernel manifold learning to cluster similar low-resolution (LR) graphs in the training sample using the weights for a set of Gaussian kernels. (2) The second step is building the brain template of LR graphs for each cluster using an iterative diffusion process with a kernel matrix P about the similarity between ROIs. (3) The last step is identifying the most K similar training LR graphs of a test subject using the multi-kernel manifold learning with three centrality metrics (degree centrality, closeness centrality, and betweenness centrality). Experimental results and comparisons with the state-of-the-art methods demonstrate that BGSR can achieve the best prediction performance, and remarkably boosted the classification accuracy using predicted HR graphs in comparison to LR, and remarkably boosted the classification accuracy using predicted HR graphs in comparison to LR graphs. We evaluated our proposed framework from ABIDE preprocessed dataset ( More details can be found at: or ![BGSR pipeline]( # Demo BGSR is coded in Python 3. In this repo, we release the BGSR source code trained and tested on a simulated heterogeneous graph data from 2 Gaussian distributions as shown in Example Results section. **Data preparation** We simulated random graph dataset from two Gaussian distributions using the function The number of graphs in class 1, the number graphs in class 2, and the number of nodes (must be >20) are manually inputted by the user when starting the demo. To train and evaluate BGSR code on other datasets, you need to provide: • A tensor of size ((n-1) × m × m) stacking the symmetric matrices of the training subjects. n denotes the total number of subjects and m denotes the number of nodes.<br/> • A vector of size (n-1) stacking the training labels.<br/> • A matrix (n × (m × m)) stacking the source HR brain graph.<br/> • Kn: the number of the most similar LR training subjects to the testing subject.<br/> • K1: the number of neighbors for applying SIMLR on clusters.<br/> • K2: the number of neighbors for applying SNF.<br/> The BGSR outputs are: • A vector of size (1 × (m × m)) vector stacking the predicted features of the testing subject. **Train and test BGSR** To evaluate our framework, we used leave-one-out cross validation strategy. To test our code, you can run: # Example Results If you set the number of samples (i.e., graphs) from class 1 to 25, from class 2 to 25, the size of each graph to 60 (nodes), K1 to 10 and K2 to 20, you will get the following outputs when running the demo with default parameter setting (alpha=0.5 and T=20 (number of iterations)): ![BGSR results]( # Acknowledgement We used the following codes: SIMLR code from SNF code from We used the following libraries: - `numpy>=1.16.6` - `networkx>=2.4` - `matplotlib>=3.1.3` - `sklearn>=0.0` Degrees, closeness and isDirected functions of this library are converted to Python: Octave networks toolbox: Functions for Pearson Correlation coded by @dfrankov in Stack Overflow. # Related references Similarity Network Fusion (SNF): Wang, B., Mezlini, A.M., Demir, F., Fiume, M., Tu, Z., Brudno, M., HaibeKains, B., Goldenberg, A., 2014. Similarity network fusion for aggregating data types on a genomic scale. [] (2014) []. Single‐cell Interpretation via Multi‐kernel LeaRning (SIMLR): Wang, B., Ramazzotti, D., De Sano, L., Zhu, J., Pierson, E., Batzoglou, S.: SIMLR: a tool for large-scale single-cell analysis by multi-kernel learning. [] (2017) []. # Please cite the following paper when using BGSR: @article{mhiri2020brain,<br/> title={Brain Graph Super-Resolution for Boosting Neurological Disorder Diagnosis using Unsupervised Multi-Topology Connectional Brain Template Learning},<br/> author={Mhiri, Islem and Khalifa, Anouar Ben and Mahjoub, Mohamed Ali and Rekik, Islem},<br/> journal={Medical Image Analysis},<br/> pages={101768},<br/> year={2020},<br/> publisher={Elsevier}<br/> }<br/> Paper link on Elsevier: Paper link on ResearchGate: # License Our code is released under MIT License (see LICENSE file for details). # Contributing We always welcome contributions to help improve BGSR and evaluate our framework on other types of graph data. If you would like to contribute, please contact Many thanks.