[论文精读]Functional Connectivity Network Analysis with Discriminative Hub Detection for Brain Disease -

论文原文：Functional Connectivity Network Analysis with Discriminative Hub Detection for Brain Disease Identification | Proceedings of the AAAI Conference on Artificial Intelligence

论文全名：Functional Connectivity Network Analysis with Discriminative Hub Detection for Brain Disease Identification

英文是纯手打的！论文原文的summarizing and paraphrasing。可能会出现难以避免的拼写错误和语法错误，若有发现欢迎评论指正！文章偏向于笔记，谨慎食用

目录

1. 省流版

1.1. 心得

1.2. 论文总结图

2. 论文逐段精读

2.1. Abstract

2.2. Introduction

2.3. Related work

2.3.1. Network-based brain disease diagnosis

2.3.2. Hub detection from networks

2.4. Methodology

2.4.1. Notations

2.4.2. Connectivity network analysis with discriminative hub detection (CNHD)

2.4.3. Optimization algorithm

2.4.4. Computational complexity

2.5. Experiments

2.5.1. Experimental results and analysis

2.6. Conclusion

3. 知识补充

3.1. L2,1范数

4. Reference

---

1. 省流版

1.1. 心得

（1）没推公式，感觉就是把特征选择和分类都集合到损失里去了？

1.2. 论文总结图

2. 论文逐段精读

2.1. Abstract

        ①Existing problems: a) extracted feature from brain connectivity might not suitable for the subsequent classifier; b) cannot exactly capture the hubs

        ②They prorposed a Connectivity Network analysis method with discriminative Hub Detection (CNHD) for brain disease diagnosis

2.2. Introduction

        ①Further explain the two existing problems

        ②They unify feature extraction and classifier together

        ③They developed an alternating direction method of multipliers (ADMM) to solve the proposed nonconvex optimization problem

        ④Datasets: 3 SCZ datasets

2.3. Related work

2.3.1. Network-based brain disease diagnosis

        ①Topolology-based methods: using graph kernal or clustering coefficient to capture features and send them to classifier

        ②Subgraph based methods: using different subgraph to represent each graph

immunodeficiency  n. [免疫] 免疫缺陷

2.3.2. Hub detection from networks

       ①Listing several methods which aim to find the hubs

2.4. Methodology

2.4.1. Notations

        ①Introducing some common notations in this paper

2.4.2. Connectivity network analysis with discriminative hub detection (CNHD)

        ①For  subjects with  ROIs/nodes each, the network set is defined as:

where  denotes the label of each participant. ,  -1 denotes the healthy samples and 1 denotes the diseased one

        ②The prediction goal:

where  is the empirical loss term,  is the regularization term. The weight matrix  is a symmetric matrix with all elements in diagonal is 0

        ③The logic loss function:

 denotes the bias term

        ④They decomposite  to sparse symmetric matrix   and  whose "columns are either entirely zeros or entirely non-zeros":

（哪里整列0或者整列非零了？）. "sparse entries of  represent edges between non-hub nodes, and the non-zero columns of are regarded as hub nodes."（我没能非常get）

        ⑤CNHD module:

where  denotes non-negative parameters

2.4.3. Optimization algorithm

        ①这，好难打字，我截图了（这个是改变上面⑤式子的约束）：

        ②They change the original loss function to:

where  denotes the Lagrange multiplier,  is the penalty parameter.

        ③Iteration steps:

2.4.4. Computational complexity

        ①Time complexity of their algorithm: 

        ②Overall algorithm:

2.5. Experiments

（1）Datasets

        ①Affiliated Nanjing Brain Hospital of Nanjing Medical University (NBH): 24 SZ and 21 NC

        ②Center for Biomedical Research Excellence (COBRE): 67 SZ and 53 NC

        ③National Taiwan University Hospital (Taiwan): 69 SZ and 62 NC

        ④Atlas: AAL-90

        ⑤FC construction: Pearson

（2）Competing methods

        ①Introducing compared methods

（3）Evaluation metric

        ①Compared metrics: ACC, SEN, SPE, balanced accuracy (BAC), positive predictive value (PPV), negative predictive value (NPV), AUC

（4）Experimental settings

        ①Validation: 5-fold cross-validation with 4 for training and 1 for testing

        ②Grid search for hyper-parameters: 

2.5.1. Experimental results and analysis

（1）Results of brain disease classification

        ①Comparison table:

（2）Analysis of detected network hubs

        ①Hubs in one fold:

        ②Visualized hubs:

（3）Convergence evaluation

        ①Iteration figure:

2.6. Conclusion

        Their model unify the feature selection and classifier, achieving a excellent performance,

3. 知识补充

3.1. L2,1范数

（1）定义：L2,1范数定义为矩阵每一行的L2范数之和

（2）举例：

4. Reference

Wang, M. et al. (2019) 'Functional Connectivity Network Analysis with Discriminative Hub Detection for Brain Disease Identification', AAAI. doi: 10.1609/aaai.v33i01.33011198