【BME 2112】W6 notes

Functional Neuroimaging: Visualizing the Working Brain

part1 quiz

1. Please briefly describe how to perform a task fMRI experiment.

2. Please describe how functional stimulation could induce localized MR signal change. In your description, you can include the changes in neuronal activity, oxygen consumption change, blood flow change, deoxyhemoglobin, and T2*.

When there is a functional stimulation, neuronal activity increasing,blood flow become quicker，oxygen consumption become faster,and then the deoxyhemoglobin decrease, the $T2\star$ signal increase, then the Localized RM signal increase.

4. What is the hemodynamic response function?

Change in MR signal related to neuronal activity (HRF:Hemodynamic Response)
Has multiple components

• Changes delayed by 1-2 sec (lags activity)
• Influx of Hb greater than needed for activity
• 5-6 sec time to peak
• Undershoot follows ~6s after peak
  

5. Please list the major steps of preprocessing fMRI data.

• DICOM to NIFTI conversion
• Slice-timing correction
• Head motion correction
• Coregistering the functional and structural data
• Normalization
• Spatial smoothing
• Temporal filtering

part2

Basic statistical tests
Regression analysis
Multiple comparisons corrections
Group analysis

Null Hypothesis $H_0$, Hypothesis $H_1$ 参考链接

The Alternative Hypothesis $H_1$ expresses outcome of interest.
$H_0:$ condition1 $\ne$ condition2 没有区别,i.e., p value 0.00001
$H_1:$ condition1 $=$ condition2 有区别
Type I error（I类错误）：犯I类错误的概率，也就是当零假设是真实的，却拒绝零假设的概率。Type II error（II类错误）：犯II类错误的概率，也就是当零假设是不真实的，却没有拒绝零假设的概率。

Massive univariate approach(大规模单变量方法）

1. T-test 基于体素的，$t=\frac{\bar{x}-\bar{y}}{{\sigma }_{xy}}=\frac{\bar{x}-\bar{y}}{\sqrt{{\sigma }_x^2+{\sigma }_y^2}}$
   $t=\frac{mea{n}_1-mea{n}_2}{StandardErroroftheMean}$

用于评估至多两个不同样本间的统计学差异。仅适用于通过少量样本（<30）,来对更大的群体做评估。

1. 比如，如果你感兴趣的变量是一个小学六年级学生的身高，那么你可以随机选取25-30个六年级学生并测量他们的身高，t test可以回答你“是否六年级学生的平均身高高于1.2米”。
2. 注意x=5处的垂线，代表了我们这个样本的均值。利用软件可以计算出垂线右侧与曲线所夹的面积（定积分），这就得到了所谓的P value(这里是0.09)。
3. 注意到我们的研究目的是探究是否平均身高显著高于1.2米，是个单向性的假设，因此我们在这里使用单边检验。
4. 如果我们在这里将显著性水平 设定为0.05，由于P值大于0.05，说明我们没有足够的正确概率来拒绝H0假设，因此接受原假设，不能证明平均身高显著高于1.2米。
   
   
   task到signal 有一个delay 不会是A那种情况

2. General Linear Model (GLM)
   没有用到volex的spatial information
   
   stimulus paradigm，age，gender
   motion parameters estimated in preprocessing steps

3. Region-of-interest approach

   1. Anatomically derived
   2. Atlas derived
   3. Functional derived
      

每一个voxel都可以计算r



T-test 只比较了两种，F-test比较了以上三种


TP（True Positive）： 正确分成A的数目，即预测为A，真值也是A，。
FP（False Positive）： 错误分成A的数目，即预测为A，真值是非A。
TN（True Negative）： 正确分成非A的数目， 即预测为非A，真值也是非A，。
FN（False Negative）： 错误分成非A的数目，即预测为非A， 真值是A。
Ture 表示进行了正确的分类，Positive和Negative表示预测值
TP：标签为正，预测为正
TN：标签为负，预测为负
FN：标签为正，预测为负，进行了错误的 负预测 TypeII error
FP：标签为负，预测为正，进行了错误的 正预测 TypeI error

Multiple comparison problem

• When we have N tests:
  Chance of making at least one type I error= $1-(1-\alpha {)}^N$ (α：一个独立测试的显著水平；N： 对比次数)
  Very large when N is large!

举个例子，对于一个10次试验的的序列，显著水平5%，FWE = ≤ 1 – (1 – .05)^10 = 0.401 ，这意味着Type I Error 发生的概率超过了40%，对于只有10次试验而言，是非常高的

• Two method：
  • FWER： Family wise error rate ：p(at least 1 FP)=FWERh
  • FDR：false discovery rate(伪发现率)：E(FP/all positives)=FDR

Family wise error rate (FWER)

• Now we have a family of hypotheses h1, h2… hn
  FWER is the probability of having at least one false positive among the family of hypotheses
  p(at least 1 FP)=FWERh
• Bonferroni correction:
  Most simple and conservative Corrected α = α/N ->FWER<= α

单步程序对每个p值进行相同的调整。这保持了整个alpha水平保持在期望值(例如…05)。该方法被称为被称为Bonferroni校正。

1. 将alpha级别除以正在运行的测试的数量，并将该alpha级别应用于每个单独的测试。例如，如果您的整体alpha水平是.05，并且您正在运行5个测试，那么每个测试的alpha水平将是.05/5=.01。
2. 在每个测试中应用新的alpha级别来查找p值。在本例中，p值必须小于或等于0.01才具有统计意义

• Not commonly used in fMRI : too conservative, false negative (type II error)

False discovery rate (FDR)

FDR controls for the proportion of positive results (i.e., discoveries) that are actually false positives.

1. Rank all p values from smallest to largest {p(i)}={p(i) :p(1)<=p(2)<=…<=p(n)}
2. Find the largest i that holds the inequality $p_i<=\frac{{\alpha }^{ii}}{n}$
3. P value of the largest index i’ p(i’) can be used as a statistical threshold

FDR的计算是根据假设检验的P-value进行校正而得到的。一般来说，FDR的计算采用Benjamini-Hochberg方法（简称BH法），计算方法如下：

1. 将所有P-value升序排列.P-value记为P，P-value的序号记为i，P-value的总数记为m
2. FDR(i)=P(i)m/i
3. 根据i的取值从大到小，依次执行FDR(i)=min{FDR(i),FDR(i+1)}
   实际上，BH法的原始算法是找到一个最大的i，满足P≤i/m*FDR阈值，此时，所有小于i的数据就都可以认为是显著的。(就是ppt中的做法)

Random field theory (RFT)

fMRI images are smooth

• intrinsic smoothness（内在光滑度）
  • Image aquisition process
  • Underlying physiological factors
• extrinsic smoothness（外在光滑度）
  • resampling during preprocessing

  如果原 2mm 用1.5倍的（采样）-》3mm

  • deliberate additional smoothing

image ≈ discretised continuous random field(离散连续随机场)
If a dataset consisting of x by y by z voxels had smoothness of V voxels, the number of independent comparisons ® would be given by $R=\frac{x\times y\times z}{V^3}$
但volex是有相关性的

Group analysis of fMRI data

• fMRI study often includes：
  • multiple subjects drawn from population
  • multiple sessions for individual subjects
• Fixed and Random effects analysis
  • Fixed-effects analysis: An analysis that assumes that the effect of the experimental manipulation has a constant effect(对所有受试者都有恒定的影响，除了随机噪声), with differences between successive observations caused by random noise.
  • Random-effects analysis: An analysis that assumes that the effects of the experimental manipulation are randomly sampled from some larger population, as when participants are drawn from the community, so that there is a distribution of effects across possible observations.（换人会有影响）
  • Mixed-effects analysis: In the context of fMRI, the common practice of modeling the effect of the experimental manipulation as stable within a participant (fixed effects) but variable across participants (random effects).