[综述笔记]A Survey on Deep Learning for Neuroimaging-Based Brain Disorder Analysis

论文网址：Frontiers | A Survey on Deep Learning for Neuroimaging-Based Brain Disorder Analysis (frontiersin.org)

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. Deep Learning

2.3.1. Feed-Forward Neural Networks

2.3.2. Stacked Auto-Encoders

2.3.3. Deep Belief Networks

2.3.4. Deep Boltzmann Machine

2.3.5. Generative Adversarial Networks

2.3.6. Convolutional Neural Networks

2.3.7. Graph Convolutional Networks

2.3.8. Recurrent Neural Networks

2.3.9. Open Source Deep Learning Library

2.4. Applications in Brain Disorder Analysis With Medical Images

2.4.1. Deep Learning for Alzheimer's Disease Analysis

2.4.2. Deep Learning for Parkinson's Disease Analysis

2.4.3. Deep Learning for Austism Spectrum Disorder Analysis

2.4.4. Deep Learning for Schizophrenia Analysis

2.5. Discussion and Future Direction

2.6. Conclusion

3. Reference List

---

1. 省流版

1.1. 心得

（1）上来直接就开模型介绍，文心吃这些东西吃多了吧

（2）我觉得不该把疾病分开诶，现在很多模型不都为了泛化而用在几个疾病数据集上吗？

（3）⭐在可解释性和数据集上给出解决办法是值得认可的

（4）哥们儿正文和discussion是一个人写的吗？？？discussion写这么好怎么正文跟

1.2. 论文总结图

2. 论文逐段精读

2.1. Abstract

        ①Structural magnetic resonance imaging (MRI), functional MRI, and positron emission tomography (PET) can all be used in neuroimage analysis

        ②Disease included: Alzheimer's disease, Parkinson's disease, Autism spectrum disorder, and Schizophrenia

2.2. Introduction

        ①Introducing medical imaging

        ②Therefore, the feature selection step is extremely important for complex medical image processing. Although sparse learning and dictionary learning have been used to extract features, their shallow architectures still limit their representation ability.

        ③The development of hardware promotes the improvement of deep learning in medical image analysis

        ④Categories of medical imaging analysis: classification, detection/localization, registration, and segmentation

        ⑤This survey mainly centers on brain disease

cardiac  adj.心脏的;心脏病的  n.心脏病患者;强心剂;健胃剂

2.3. Deep Learning

2.3.1. Feed-Forward Neural Networks

        ①The function of FFNN:

where the  is the input vector,  is the output;

superscript denotes layer index,  is the number of hidden units;

 and  are bias terms of input layer hidden layer respectively;

 and  denote non-linear activation function;

 represents parameter set

        ②Sketch map of (A) single and (B) multi layer neural networks:

2.3.2. Stacked Auto-Encoders

        ①Auto-encoder (AE), namely so called auto-associator, possesses the ability of encoding and decoding

        ②AE can be stacked as stacked auto-encoders (SAE) with better performance

        ③Sketch map of SAE:

where the blue and red dot boxes are encoder and decoder respectively

        ④To avoid being trapped in local optimal solution, SAE applies layer-wise pretraining methods

2.3.3. Deep Belief Networks

        ①By stacking multiple restricted Bolztman machines (RBMs), the Deep Belief Network (DBN) is constructed

        ②The joint distribution of DBN:

where  denotes visible units and  denotes  hidden layers

        ③Sketch map of (A) DBN and (B) DBM:

where the double-headed arrow denotes undirected connection and the single-headed arrow denotes directed connection

2.3.4. Deep Boltzmann Machine

        ①Futher stacking RBMs can get Deep Boltzmann Machine (DBM):

2.3.5. Generative Adversarial Networks

        ①Simultaneously including generator  and discriminator , Generative Adversarial Networks (GANs) achieves the task of training models with a small number of labeled samples:

        ②The framework of GAN:

2.3.6. Convolutional Neural Networks

        ①The framework of convolutional neural network (CNN):

2.3.7. Graph Convolutional Networks

        ①The framework of Graph Convolutional Networks (GCN):

which includes spectral-based and spatial-based methods

2.3.8. Recurrent Neural Networks

        ①As the extension of FFNN, recurrent neural network (RNN) ia able to learn features and long-term dependencies from sequential and time-series data

        ②Framework of (A) long-short-term memory (LSTM) and (B) Gated Recurrent Unit (GRU):

2.3.9. Open Source Deep Learning Library

        ①Some toolkits of deep learning:

2.4. Applications in Brain Disorder Analysis With Medical Images

2.4.1. Deep Learning for Alzheimer's Disease Analysis

        ①Introducing the Alzheimer's Disease Neuroimaging Initiative (ADNI) and the classification method of patients

        ②Enumerating DGM based and CNN based methods, 2D CNN and 3D CNN

        ③Articles which applying DL in AD detection:

        ④Classification performance of these articles:

        ⑤Articles that applying DL to predict MCI:

        ⑥Prediction performance of artivles above:

2.4.2. Deep Learning for Parkinson's Disease Analysis

        ①Dataset example: Parkinson's Progression Markers Initiative (PPMI)

        ②Exampling some DL works on PD diagnosis

        ③Articles which applying DL in PD detection:

2.4.3. Deep Learning for Austism Spectrum Disorder Analysis

        ①Dataset: ABIDE I/II

        ②Particularizing AE/CNN/RNN based methods

        ③Articles that applying DL to ASD diagnosis:

2.4.4. Deep Learning for Schizophrenia Analysis

        ①There is no widely used SZ neuroimaging dataset available currently

        ②Dataset from challenge: The MLSP2014 (Machine Learning for Signal Processing) SZ classification challenge, with 75 NC and 69 SZ

        ③Articles which applying DL in SZ detection:

2.5. Discussion and Future Direction

        ①Hyper-parameters of DL: 

model optimization parameters	the optimization method, learning rate, and batch sizes, etc.
network structure parameters	number of hidden layers and units, dropout rate, activation function, etc.

        ②Optimization of hyper-parameters: 

manual	grid search and random search
automatic	Bayesian Optimization

        ③Deep learning still faces the challenges of weak interpretability, limited multi-modality and limited data in imaging studies

2.6. Conclusion

        Medicine and computers will inevitably merge

3. Reference List

Zhang, L. et al. (2020) 'A Survey on Deep Learning for Neuroimaging-Based Brain Disorder Analysis', Front Neurosci. doi: 10.3389/fnins.2020.00779