大家好,请帮忙分析这篇文章的数据是怎么处理的

Submitted by nutrition-li on
最近在看一篇文章,有些数据分析方面的问题搞不清楚,如果有人处理过类似的数据,请不吝赐教。谢谢!
MR Image Acquisition
Images were acquired with a multislice, singleshot EPI sequence to achieve whole brain coverage: 40 axial slices, TR=5 seconds, TE =35 illiseconds,
3.5-mm slice thickness, in-plane resolution 1.8 mm by 1.8 mm at the Magnetic Resonance Imaging Facility, Salford Royal NHS Foundation Trust, Salford. A SENSE 8-channel head coil was used for radio frequency transmission and reception.

Image and Statistical Analysis

The physiologic imaging data were analyzed with the use of the Statistical Parametric Mapping (SPM5) with the pseudo-block pharmacologic
MRI analysis technique. Images were realigned to correct for motion and were spatially normalized and smoothed to facilitate intersubject averaging. Four sets of scans were excluded from study 1 (3 because of excessive movement by the participant in the baseline period of 1
of the 2 scans, and 1 because of a lack of brain stem coverage in the participant’s images), and 2 sets of scans were excluded from study 2 (both because of excessive movement in the baseline of 1 of the 4 scans as stated above).
First-level analysis was performed on each participant for each study condition in the following ways. The preinfusion and postinfusion scans were split into time bins; the postinfusion scans were divided into 15 consecutive 2-minute time bins (T1–T15) and the 24 scans from the 2 minutes immediately before the lipid/saline infusion formed the baseline time bin (T0). In each participant, the signal averages for the 15 postinfusion time bins (T1–T15) were compared with the baseline average (T0), using regression analysis for each condition (lipid, saline, lipid-dex, saline-dex). A subtraction between conditions resulted in 15 first-level images corresponding to the BOLD signal change from baseline between conditions  in each successive postinfusion time bin for each participant.


我有疑问的地方已经用黄色标出了。1、time bins这个概念怎么理解,为什么要把scans分成time bins?
2、文章中统计分析用的是BOLD signal,请问BOLD signal的数值时怎么得到的?上图就是文章中的图表,我想知道BOLD的改变是怎么计算出来的。
我刚刚开始学习fMRI,有许多问题不是很懂,也不知道我是否把问题说明白了,如果没说明白请大家留言,我继续补充。
感谢大家!

下面是我的理解,供参考。

注射之前扫描了2分钟,TR为5秒,共得到24幅全脑,也即24 scans,将这24个scans平均,得到T0,即一个基线状态的bin。照此方法,注射之后共有15个bin (T1-T15)。(T1-T0)/T0得到图中的第T1个bin百分率。4种情况,每种情况得到15个这样的变化率。所谓AUC (area under curve),估计就是将这15个bin的变化率相加。但你似乎没提供图的说明,我看不明白左上图是来自于一个人还是所有人,如果是后者,应该提供标准差。右上图的四个bar似乎应该对应四种状态,并且是来自所有人,因为bar上有标准差,但未明确注明每个bar是哪种情况,根据图下面的+ - 号,推测与左图的顺序是一样的。

谢谢臧老师指导!
在做second-level analysis的时候,文章中是这样写的“To determine whether robust statistically significant increments in the BOLD signal change across participants occurred after the infusion of lipid in both studies, a second-level conjunction analysis was carried out with a 2*15 full-factorial analysis of variance (ANOVA) with study as an independent factor and time bin (representing the lipid–saline subtraction across time) as a repeated measures nonindependent factor.”
您能帮我简单解释一下conjunction analysis是什么意思吗?它与full-factorial analysis 是一个意思吗?
这种分析在SPM中怎样才能实现呢?谢谢
 

这个问题我没什么把握,共有四种conditions,每种有15个bins,前面基于AUC的统计方法比较清楚,尽管上次并未介绍是如何进行统计的。我不明白为什么是2*15,似乎是把四种conditions合并为两个,即所谓conjunction analysis? 请其他朋友帮忙回答。

臧老师,您好!
为了让您和论坛里的其它朋友更好的讨论这个问题,我已经将这篇文献的全文上传到网上了。大家可以通过以下地址下载。
下载地址1:http://g.zhubajie.com/urllink.php?id=10372768hvaz7641b8dvvvxt
下载地址2:http://d.namipan.com/d/543709ff895fac783173c649a805d9e806e18a178bbb0d00
这篇文章包含两个study,在第一个study中,作者为了观察到灌胃C12脂肪酸之后,大脑激活的时间和空间形式(spatiotemporal pattern),对每个受试者都进行了两次40min的扫描。第一次扫描时在10min的基线期之后,灌胃生理盐水(作为C12的对照);第二次扫描时在10min的基线期之后,灌胃C12脂肪酸。灌注时间均为2min。
在进行这部分数据处理时,首先对每个被试进行1st level analysis,正如臧老师在前面讲到的,将灌胃前和灌胃后划分为16个时间区间(T0、T1、T2......T15),每个区间2min。接下来的分析我就不明白了。
First level analysis:
In each participant, the signal averages for the 15 postinfusion time bins (T1–T15) were compared with the baseline average
(T0), using regression analysis for each condition (lipid, saline,). 这里的regression analysis应该怎么理解,这一步的计算对整体的分析起什么作用?

A subtraction between conditions resulted in 15 first-level images corresponding to the BOLD signal change from baseline between conditions
(lipid vs saline in study 1 ) in each successive postinfusion time bin for each participant. 这里指的是用什么数据做的减法?

Second level analysis:
To determine whether robust statistically significant increments in the BOLD signal change across participants occurred after the infusion of lipid in both
studies, a second-level conjunction analysis was carried out with a 2 *15 full-factorial analysis of variance (ANOVA) with study as an independent factor and time bin (representing the lipid–saline subtraction across time) as a repeated measures nonindependent factor.  应该怎么理解conjunction analysis,是full-factorial analysis of variance 的同义词吗?这里用到的数据时1st level做减法之后产生的数据吗?

This conjunction analysis was thresholded at P for false discovery rate corrected <.025.  FDR的P值是凭经验呢,还是可以通过计算得到?

For display purposes the conjunction analysis has been thresholded at Punc <.005 in Figure 1.  这里的P值是怎么选的呢?

Brain areas showing significant increases in BOLD signal caused by lipid infusion in both studies were identified with the WFU PickAtlas Tool Version 2.0.

感谢臧老师、严师兄和论坛里各位朋友不吝指教!

我认为这种讨论方式非常好!不过,恕我暂时抽不出时间仔细读全文,只能部分回答你的问题。有些问题,也超出了我的能力范围。
In each participant, the signal averages for the 15 postinfusion time bins (T1–T15) were compared with the baseline average (T0), using regression analysis for each condition (lipid, saline,).”如果是通常意义的GLM,介绍回归时,通常不会与average相混淆,尽管两者有某种程度的等价。如果指的是通常意义的GLM,一般会介绍采用了什么样的HRF,当然,不与HRF卷积,理论上也应该可以直接进行回归分析。
 
“A subtraction between conditions resulted in 15 first-level images corresponding to the BOLD signal change from baseline between conditions (lipid vs saline in study 1 ) in each successive postinfusion time bin for each participant.”这儿,作者再次明确提到了减法,但没提供细节:对于lipid,如果用T1-T15分别与T0相减,得到15个图,对于saline也是如此。但"between conditions (lipid vs saline in study 1 )”,明确指两种condition之间的比较,或许是两种condition之内的差值之间再相减?
 
Second level, conjunction analysis我不太理解。
 
校正之后P值选择,比较任意,与普通的P值一样,一般<0.05都是合理的。
“For display purposes the conjunction analysis has been thresholded at Punc <.005 in Figure 1.” FDR校正之后的0.05与不校正的0.005并没有肯定的对应关系,所以,我认为,作者这样做不严谨,显示时应该显示校正后的结果。
Forums