[论文精读]FCNet: A Convolutional Neural Network for Calculating Functional Connectivity from Functional

论文全名：FCNet: A Convolutional Neural Network for Calculating Functional Connectivity from Functional MRI

论文网址：FCNet: A Convolutional Neural Network for Calculating Functional Connectivity from Functional MRI | 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. Method

2.3.1. Data and Preprocessing

2.3.2. Functional Connectivity Through FCNet

2.3.3. Feature Selection and Classification

2.4. Experiments and Results

2.5. Conclusion

3. Reference List

---

1. 省流版

1.1. 心得

（1）FC时间上相关的啊，意思空间相关不能叫FC咯？确实，想想也不是没道理，fmri本来就是功能影像（时间），smri才是结构（空间）

（2）很久没看到语言上这么奇怪的论文了，怎么感觉和我不是一个世界的人

（3）以前古早的可视化确实做的不是那么好，可以添加颜色捏

1.2. 论文总结图

2. 论文逐段精读

2.1. Abstract

        ①Traditional correlation and distance measures are essential for machine learning in fMRI technique. However, these method may not caputure precise time series feature

        ②Thus, the authors put forward FCNet, which extract functional connectivity (FC) from time signals directly

        ③They aim to classify ADHD and HC

2.2. Introduction

        ①FC presents the pair-wise temporal correlation of each brain regions

        ②⭐The approaches of FC constructions are correlation, clustering, graph based etc...

        ③The overall framework of FCNet:

(a) Train FCNet by data, which generated by generator;

(b) Generating FC by FCNet, selecting features and connecting them with phenotypic data. Then training SVM

(c) Classifying by SVM

（我天这段话..怎么...有那么点...绕口...）

2.3. Method

2.3.1. Data and Preprocessing

        ①Dataset: ADHD-200

        ②Sites: NeuroImage (NI), New York University Medical Center (NYU) and Peking University (Peking)

        ③Atlas: automated anatomical labelling atlas (AAL) 90

2.3.2. Functional Connectivity Through FCNet

        ①FCNet is a deep-network architecture

        ②The architecture of FCNet:

（1）The Feature Extractor Network

        ①Activate function: Leaky Rectified Linear Unit (ReLU), cuz it converges faster than ReLU

        ②Length of time-series signal: 172

        ③Pool length in pooling layer: 2

        ④Kernal size: 3

        ⑤Filters of C 1-5: 32, 64, 96, 64, 64

        ⑥The size of the last fully connected layer (purple blocks): 32

（2）The Similarity Measure Network

        ①Extracting correlation by neural network

        ②The essence of building relevance is to utilize similarity

        ③The last layer contains a Softmax

（3）Coupled Architecture with Shared Parameters

        ①They share the parameters in the coupled structure（呃呃为什么要写得这么抽象，所以意思是图上两根蓝线是一个人的两个ROI吗？）

（4）Data Generator for Training FCNet

        ①⭐They need similar regions, which means functionally connected and dissimilar regions, which means not connected functionally. Labelling similar one with 1, dissimilar one with 0（救命...这是什么）

        ②Labelling 1 to ROIs with the same communities and 0 with dissimilar communities

        ③The specific algorithm:

（5）Training of FCNet

        ①Loss function: cross-entropy

where  denotes the number of training samples, ,  denotes the predicted result

2.3.3. Feature Selection and Classification

        ①Feature selection: Elastic Net (EN) based

        ②EN combines  and  penalty that  brings variable selection and continuous shrinkage and  encourages grouped selection of features

        ③Defining  as the vector of subjects  and , the cost function will be:

where  are weights;

 needs to be calculated

2.4. Experiments and Results

        ①Classes: HC and ADHD (combined with ADHD combined, ADHD hyperactive-impulsive and ADHD inattentive)

        ②The model proposed by the authors outperforms all other models in the competition. And for datasets from different sites, the parameters in their frameworks remain unchanged.

        ③Comparisons in competition:

        ④Validation of phenotype data usage:

        ⑤Visualization of FC comparison between HC and ADHD on Peking site:

2.5. Conclusion

        阐述了一下它究竟有多么复杂

3. Reference List

Riaz, A. et al. (2017) 'FCNet: A Convolutional Neural Network for Calculating Functional Connectivity from Functional MRI', Medical Image Computing and Computer Assisted Intervention Society (MICCAI): Connectomics in NeuroImaging (CNI). doi: 10.1007/978-3-319-67159-8_9