[论文精读]Intelligence Quotient Scores Prediction in rs-fMRI via Graph Convolutional Regression Network

论文网址：Intelligence Quotient Scores Prediction in rs-fMRI via Graph Convolutional Regression Network | SpringerLink

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. Methodology

2.3.1. Pipeline

2.3.2. Proposed Network

2.4. Experimental Results

2.4.1. Experimental Settings

2.4.2. Performance Evaluation

2.4.3. Biomarker Detection

2.5. Conclusion

3. 知识补充

3.1. Poly learning rate

4. Reference List

---

1. 省流版

1.1. 心得

（1）用自闭症数据集来预测智商？？好吧（我真的想说这后面也没有圆回来！！！我还以为有什么惊为天人的发现呢！）

（2）从前面看是好简单的一个网络啊！后面看也没有很难

（3）这个随机选样本，hhhhh

（4）你有什么拓扑结构？？？我怎么没get到。提取的特征就是高级拓扑特征？

（5）宝宝你是一个...

1.2. 论文总结图

2. 论文逐段精读

2.1. Abstract

        ①Due to the complexity of brain connections and the change of topological structure in brain, IQ predictions always get inaccurate results

        ②They put forward a model including attention branch and global branch

2.2. Introduction

        ①There is no relevant topic of IQ prediction in ASD datasets

        ②Machine learning separate feature selection and regression. However, it is hard to obrain accurate advanced topological features from brain networks

        ③They proposed a Graph Convolutional Regression Network (GCR-Net) for IQ prediction

engrain  v.使根深蒂固，深透，确立  adj.根深蒂固的

2.3. Methodology

2.3.1. Pipeline

        ①The schematic of GCR-Net

        ②Atlas: Automated Anatomical Labeling (AAL) with 116 ROIs

        ③Functional connectivity matrix : calculated by Pearson correlation 

        ④The brain graph , where  are ROIs

2.3.2. Proposed Network

        ①Their overall framework:

where  denotes the node features

        ②The ConvNet includes a graph convolution (GC) layer and a dropout layer

（1）Attention branch

        ①The operation of GC:

where  represents the degree matrix of  and they set （等等，为什么我这里怎么也理解不了，这A是个啥玩意儿啊？）;

 denotes node features

        ②In the self attention module, the attention score  can be calculated by:

        ③The pooling module contains matrix multiplication  and maximum pooling. After pooling, it generates prediction 

（2）Global branch

        ①The GC operation:

        ②Transposing  and regrding it as a global (personal) feature

        ③MLP 

（3）Finally, fusing  and  with element-wise mean operation then giving the prediction

2.4. Experimental Results

2.4.1. Experimental Settings

（1）Settings

        ①Optimizer: Adam

        ②Learning rate: 0.001

        ③Weight decay: 0.0005

        ④⭐Poly learning rate: 0.9 power

        ⑤Maximum epoch: 100

        ⑥Loss function: mean square error (MSE)

        ⑦Hyper-parameters: 

（2）Dataset

        ①Autism Brain Imaging Data Exchange (ABIDE) adopted

        ②Preprocessing pipelines: Connectome Computation System (CCS), Configurable Pipeline for the Analysis of Connectomes (CPAC), Data Processing Assistant for Resting-State fMRI (DPARSF), Neuroimaging Analysis Kit (NIAK)

        ③⭐Samples: randomly choosing 226 NT and 202 ASD

（3）Evaluation Metrics

        ①Cross validation: 3 fold, 2 for training and 1 for test

        ②Runs: mean of 3

        ③mean absolute error (MAE): 

        ④root mean squared error (RMSE): 

where  is the predicted answer and  is the true label,  denotes the number of patients

2.4.2. Performance Evaluation

        ①Regression comparison in ASD:

        ②Regression comparison in NT:

2.4.3. Biomarker Detection

        ①5 most influential ROIs for IQ on ASD:

        ②5 most influential ROIs for IQ on NT:

        ③Analysing the similarity between predicted associated brain regions and medical research

episodic  adj.不定期的;情节性的;偶尔发生的;有许多片段的;由松散片段组成的

2.5. Conclusion

        No need for that.

3. 知识补充

3.1. Poly learning rate

参考学习1：Pytorch几种常用的学习率调整方式_pytorch poly-CSDN博客

参考学习2：pytorch动态调整学习率之Poly策略_poly学习率-CSDN博客

4. Reference List

Zhang H. et al. (2022) 'Intelligence Quotient Scores Prediction in rs-fMRI via Graph Convolutional Regression Network', CAAI International Conference on Artificial Intelligence, pp. 477-488. doi: Intelligence Quotient Scores Prediction in rs-fMRI via Graph Convolutional Regression Network | SpringerLink