①Sample filtering: excluding which movement higher than 0.2 mm, then 883 left; 进一步检查了构成DMN的18个大脑区域的BOLD时间序列，删除了其中一个 BOLD 时间序列为零的个体数据。Finally left 871 participants with the portion between ASD and TD individuals is approximately 1.18

2.3.3. Feature Extraction

①They include demographic information and patient movement information

（1）Static Functional Connectivity

①Atlas: CC200

②Selected brain regions: 153 from 18 DMN regions

③sFC: calculated by Pearson using pairs of BOLD time series:

$\mathrm{sFC}=\rho =\frac{\mathrm{Cov}\left(\mathrm{x},\mathrm{y}\right)}{\sqrt{\mathrm{Var}\left(\mathrm{x}\right)\mathrm{Var}\left(\mathrm{y}\right)}}$

（2）Dynamic Functional Connectivity

①dFC: calculated by window slicing with 65 length, 移动一个重复时间（TR）点，该点可以是几个时间点，具体取决于聚合采集时间。153 features as well.

（3）Demographics

①Applied 2 demographic data: age and handedness（那为什么要把性别放进来？？）（根据后面特征数量的计算，这里确实没有加入性别）

ambidextrous adj.左右手都灵巧的；左右开弓的

（4）Acquisition Parameters

①9 parameters: "field of view (FOV, 3 features—one for each dimension, x, y, and z), repetition time (TR, 1 feature), echo time (TE, 1 feature), voxel size (3 features corresponding to space dimensions), and number of slices (1 feature)"

（5）Motion Parameters

①⭐Head motion parameter: the parameters are calculated by "framewise displacement (FD) provided from each datacenter (2 features corresponding to FD mean value and percentage of slices with FD over 0.2 mm)"

②Rate of change in the BOLD signal (DVARS): calculated by "temporal derivative of time courses and the variance over voxels of root mean square"

2.3.4. Feature Selection and Biomarker Assessment

①Workflow of proposed method:

②⭐作者这里说有473（153 of sFC+306 of dFC+2 of demographics+9 of acqusition+3 of motion）个特征/人，但是一共有871个人。作者引用(Ambroise & McLachlan, 2002)说明特征参数最好是被试的三分之一不然容易过拟合（但是有点早了，也不能说完全可靠吧），因此需要引入如下的特征选择方式：

③Feature selection (FS) method: recursive feature elimination with correlation bias reduction (RFE-CBR)（类似SVM每次删除最不重要的以及除去高度相关的特征）

④Cross validation: 10 fold

2.3.5. Classification

①Classifier: SVM

②Optimize: minimizing the $error=1-AUC$

③Set ratio: 90% for training and 10% for testing

④Maximum epoch: 50

2.4. Results

2.4.1. Classification Performance Results

①Performance: 76.63% ACC, 78.63% SEN, 74.27% SPE and 82.74% AUC with 136 features（不知道为什么ABIDE数据集的AUC都普遍比ACC高）

②⭐作者觉得ASD患者性别分布悬殊所以性别也是潜在的生物标志物

③AUC figure:

④136 feature included (dFC most):

included

a) dFC characteristics (47 and 45 out of the 136 total features for mean dFC and variance dFC, respectively),

b) sFC incorporating 43 (out of the 136)

c) acqusition: echo time (TE)

excluded

demographics, motion, and the rest of the acquisition features (FOV, voxel size, and repetition time)

⑤Discriminative features:

where green line denotes sFC, blue denotes variance dFC, red denotes mean value dFC

2.4.2. Analysis of Selected Biomarkers

①Lists discriminative brain regions

2.4.3. Functional Connectivity Features Validation

①They employed Wilcoxon test on ASD and TD, finding that there are significant differences between 19 features. These significant differences includesFC and mean dFC

2.4.4. Evaluation of the Adopted Framework and Relationship to Prior Work

①Performance comparison table:

2.5. Discussion

2.5.1. Classification Performance

①作者觉得不同的数据选择让方法之间无法比较，我的评价是得自己去复现别人的代码在自己选的数据上。这里作者确实可能没去做这个工作量

②作者觉得必须筛选一些不好的数据，其实这个还是备受争议啦，毕竟人家ABIDE采的时候就筛选过了

2.5.2. Informative Biomarkers

①Introducing some medical speculation on ASD

plethora n.过多；过剩；过量 excitatory adj.兴奋的；有刺激性的，兴奋的

oscillation n.振动；摆动；振幅；摇摆；(情感、行为、思想的)摇摆不定，变化无常，犹豫不定；浮动；一次波动

2.6. Conclusions

Their model got the most highest performance

3. 知识补充

3.1. recursive feature elimination with correlation bias reduction (RFE-CBR)

（1）论文：https://www.sciencedirect.com/science/article/pii/S0925400515001872

（2）参考学习：【数据处理系列】深入理解递归特征消除法（RFE）：基于Python的应用-CSDN博客

4. Reference

Ambroise, C. & McLachlan, G.J. (2002) ‘Selection bias in gene extraction on the basis of microarray gene-expression data’, Proc. Natl. Acad. Sci. 99: 6562–6566. doi: https://doi.org/10.1073/pnas.102102699

Karampasi, A. S. et al. (2024) 'Informative Biomarkers for Autism Spectrum Disorder Diagnosis in Functional Magnetic Resonance Imaging Data on the Default Mode Network', Appl. Sci., 11. doi: Applied Sciences | Free Full-Text | Informative Biomarkers for Autism Spectrum Disorder Diagnosis in Functional Magnetic Resonance Imaging Data on the Default Mode Network