Draft

<Readme: 檔案說明> Homer2 Slides + 測試檔案 分析工具的資料

Coding for Matlab Clinical Data:

1. head model 2. pathlenge 3. MBLL mua a (消光係數)

newdata: 三週新收案的解壓縮檔 (已經輸出xls file) extract_newdata: 前述輸出的xls+BG demo_file: (20210506) 預先建好的資料夾(for 5 subjects)/.nirs格式mat變數、cfg

input_test: 測試環境，

可能用不到的方法(?):

• GLM-BCI-master
• tCCA_model

※ Time series 可能不是2310 = 77 x 30 (視原始檔案和Sampling Rate而定) =>用matlab生成數列比較快

• Correct the issue of datetime when launching Homer3: "func_correction_for_Homer3"

Y = 3.5126 X+532.49 % Wavelets Motion Corrections (para = 1.5) % Bandpass filter (High pass filter = 0.01; Low pass filter = 0.10)

CBSI: Baseline of HbT (if motion is too much, can be bad for filter)

PCA: for motion artifact really huge (> 5 SD)

% Manual Removal of Motion Segments

Spline

Wavelet

Do before MBLL or after? (Actually, we don't know!) ==============

Mayer Waves: 0.1 Hz (variable) / sitting or standing is higher than laying (that's why fMRI can ignore this)(?)

=> The enigma of Mayer waves:Facts and models.

Cardiac Signal (~70bpm/ 1.16Hz)

Respiration (~0.2Hz in Adult)

Drift : Amplitude Voltage change

Frequency Filtering: Most GLM stimuli work best at the 0.1-0.3 Hz range

Resting-state hemodynamics are dominated by slow-5 (110^-2 ~ 2.710^-2) slow-4(2.710^-2~7.310^-2) =>Lower

=============== <UCLA_Preprocessing Pipeline>

1. Identify windows of motion artifact
2. PCA motion artifact correction
3. MBLL conversion (OD2Conc)
4. Frequency filtering (bandpass filter, 0.005Hz-0.5Hz or a slower low-pass cutoff if you know your GLM presentation rate)

David: Yeah, when using the GLM there really is no need for using a bandpass filter before the GLM (??)

5. Z-score

Before, we also do automatic "bad channel" rejection - throw out channels that are mostly noise and have white noise-shaped frequency spectrum (=> Power frequency spectrum)

?: actually use Quartile Coefficient of Dispersion: (Q1-Q3)/(Q1+Q3)