[论文精读]Community-Aware Transformer for Autism Prediction in fMRI Connectome

论文网址：[2307.10181] Community-Aware Transformer for Autism Prediction in fMRI Connectome (arxiv.org)

论文代码：GitHub - ubc-tea/Com-BrainTF: The official Pytorch implementation of paper "Community-Aware Transformer for Autism Prediction in fMRI Connectome" accepted by MICCAI 2023

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！

1. 省流版

1.1. 心得

（1）我超，开篇自闭症是lifelong疾病。搜了搜是真的啊，玉玉可以治愈但是自闭症不太行，为啥，太神奇了。我还没有见过自闭症的

1.2. 论文总结图

2. 论文逐段精读

2.1. Abstract

①Treating each ROI equally will overlook the social relationships between them. Thus, the authors put forward Com-BrainTF model to learn local and global presentations

②They share the parameters between different communities but provide specific token for each community

2.2. Introduction

①ASD patients perform abnormal in default mode network (DMN) and are influenced by the significant change of dorsal attention network (DAN) and DMN

②Com-BrainTF contains a hierarchical transformer to learn community embedding and a local transformer to aggregate the whole information of brain

③Sharing the local transformer parameters can avoid over-parameterization

2.3. Method

2.3.1. Overview

（1）Problem Definition

①They adopt Pearson correlation coefficients methods to obrain functional connectivity matrices

②Then divide $N$ ROIs to $K$ communities $\{X_{1},X_{2},\ldots,X_{K}\},X_k\in\mathbb{R}^{N_k\times N}$

③The learned embedding $H=[H_{1},\ldots,H_{k}],H_k\in\mathbb{R}^{N_k\times N}\mapsto Z_{L}\in\mathbb{R}^{N\times N}$

④Next, the following pooling layer and MPLs predict the labels

（2）Overview of our Pipeline

①They provide a local transformer, a global transformer and a pooling layer in their local-global transformer architecture

②The overall framework

2.3.2. Local-global transformer encoder

①With the input FC, the learned node feature matrix $H_i$ can be calculated by $H_i=(\|_{m=1}^Mh^m)W_O$

②In transformer encoder module,

$h^m=\text{softmax}\bigg(\frac{Q^m(K^m)^T}{\sqrt{d_k^m}}\bigg)V^m$

where $Q^{m}=W_{Q}X_{i}^{\prime},K^{m}=W_{K}X_{i}^{\prime},V^{m}=W_{V}X_{i}^{\prime},X_{i}^{\prime}=[p_{i},X_{i}]$ ,

$M$ is the number of heads

（1）Local Transformer

①They apply same local transformer for all the input, but use unique learnable tokens $\{p_1,p_2,...,p_k\},p_i\in\mathbb{R}^{1\times N}$ :

$p_i',H_i=\text{LocalTransformer}([p_i,X_i])\text{where},i\in[1,2...K]$

（2）Global Transformer

①The global operation is:

$p_{global}=\text{MLP (Concat }(p_1^{'},p_2^{'}\ldots p_K^{'}))$

$H_{global}=\text{Concat}(H_1,H_2,\ldots,H_K)$

$p^{'},Z_L=\text{GlobalTransformer}([p_{global},H_{global}])$

2.3.3. Graph Readout Layer

①They aggregate node embedding by OCRead.

②The graph level embedding $Z_G$ is calculated by $Z_{G}=A^{\top}Z^{L}$ , where $A\in\mathbb{R}^{K\times N}$ is a learnable assignment matrix computed by OCRead layer

③Afterwards, flattening $Z_G$ and put it in MLP for final prediction

④Loss: CrossEntropy (CE) loss

2.4. Experiments

2.4.1. Datasets and Experimental Settings

（1）ABIDE

（2）Experimental Settings

2.4.2. Quantitative and Qualitative Results

2.4.3. Ablation studies

（1）Input: node features vs. class tokens of local transformers

（2）Output: Cross Entropy loss on the learned node features vs. prompt token

2.5. Conclusion

2.6. Supplementary Materials

2.6.1. Variations on the Number of Prompts

2.6.2. Attention Scores of ASD vs. HC in Comparison between Com-BrainTF (ours) and BNT (baseline)

2.6.3. Decoded Functional Group Differences of ASD vs. HC

3. 知识补充

4. Reference List

Bannadabhavi A. et al. (2023) 'Community-Aware Transformer for Autism Prediction in fMRI Connectome', 26th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2023), doi: https://doi.org/10.48550/arXiv.2307.10181

本文来自互联网用户投稿，该文观点仅代表作者本人，不代表本站立场。本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如若转载，请注明出处：http://xiahunao.cn/news/2778285.html

如若内容造成侵权/违法违规/事实不符，请联系瞎胡闹网进行投诉反馈，一经查实，立即删除！