Converting the overnight hypnogram and handling abnormal spectrogram values

[qaqis95]

Excellent work!

1. Extracting sleep EEG data from the sleep EDF file and converting it to the npz format, I obtained very low values for the first five elements. What could be the cause of this outcome?
   F4-M1: [-3.05180438e-08 -3.05180438e-08 -3.05180438e-08 -3.05180438e-08
   -3.05180438e-08]
   C4-M1: [-3.05180438e-08 -3.05180438e-08 -3.05180438e-08 -3.05180438e-08
   -3.05180438e-08]
   O2-M1: [-3.05180438e-08 -3.05180438e-08 -3.05180438e-08 -3.05180438e-08
   -3.05180438e-08]
   F3-M2: [-3.05180438e-08 -3.05180438e-08 -3.05180438e-08 -3.05180438e-08
   -3.05180438e-08]
   C3-M2: [-3.05180438e-08 -3.05180438e-08 -3.05180438e-08 -3.05180438e-08
   -3.05180438e-08]
   O1-M2: [-3.05180438e-08 -3.05180438e-08 -3.05180438e-08 -3.05180438e-08
   -3.05180438e-08]

2. The spectrogram values obtained from my overnight sleep data are lower compared to the example data. This raises the question of what could be causing this discrepancy.

3. I would like to replicate the "Artifact Rejection" section.
   I initially converted the overnight hypnogram from txt format to npz format, which resulted in 1308 arrays. However, when converting the entire overnight EEG to npz format, it resulted in 10049536 arrays, causing a data mismatch. How can this issue be resolved?
   Below is my code.

Stage2 = np.loadtxt('2.txt')
print(Stage2)
np.savez('Stage2.npz', data=Stage2)
f = np.load('data.npz')
data= f['data']
chan = ['F4-M1', 'C4-M1', 'O2-M1', 'F3-M2', 'C3-M2', 'O1-M2']
sf = 256.
hypno = np.load('Stage2.npz').get('data')
times = np.arange(data.size) / sf
print(data.shape)
print(chan)
print(np.round(data[:, 0:5], 3))
data.shape, hypno.shape

---output--
[0. 0. 0. ... 0. 0. 0.]
(6, 10049536)
['F4-M1', 'C4-M1', 'O2-M1', 'F3-M2', 'C3-M2', 'O1-M2']
[[-0. -0. -0. -0. -0.]
[-0. -0. -0. -0. -0.]
[-0. -0. -0. -0. -0.]
[-0. -0. -0. -0. -0.]
[-0. -0. -0. -0. -0.]
[-0. -0. -0. -0. -0.]]
((6, 10049536), (1308,))

[raphaelvallat]

Hi,

1. Are you using MNE to load your EDF data? MNE automatically converts the data to Volts, so you need to multiply by 1e6 if you want the data to be in micro-Volts.

2. Do you mean lower in terms of the log-power unit, in dB / Hz? The absolute amplitude of the EEG is impacted by age, sex, skull thickness, as well as the PSG system and electrode location, so would not worry too much about this.

3. You need to upsample your hypnogram to the sampling frequency of the EEG data, by using this function: https://raphaelvallat.com/yasa/build/html/generated/yasa.hypno_upsample_to_data.html as explained in this section of the quickstart

[raphaelvallat]

Where do you see the 620 on the above screenshot?

represents the number of so-spindle couplings detected by the F4-M1 channel.

No, when you run yasa.sw_detect(coupling=True), the coupling values are not based on the spindles detection. Instead, for each slow-waves, the coupling with the filtered sigma band is calculated.

[qaqis95]

Thank you so much for your thorough explanation! @raphaelvallat
I think I have understood the 'yasa.sw_detect(coupling=True)' .

1. The value 620 represents the number of rows in the screenshot's bottom-left corner. I think this value should correspond to the count of sigma-so couplings. Is that correct?
   

2. Regarding the so-spindle couplings, what method would you recommend? I also noticed that 'find_cooccurring_spindles' does not seem to provide information on the phase and strength of the so-spindle coupling.

3. Currently, I have completed data analysis for 20 participants by Yasa. The sleep stages were manually classified. One of the participants had 126 minutes of N2 sleep and 30 minutes of N3 sleep, but very few or even no spindles were detected. What do you think could be the possible reasons for this?
   

Thanks!

[raphaelvallat]

Hi,

1. No, that value represents the total number of detected slow-waves. Spindles are not detected here.
2. I do not recommend using the find_cooccurring_spindles to calculate coupling. Please see this notebook for more details: https://github.com/raphaelvallat/yasa/blob/master/notebooks/12_SO-sigma_coupling.ipynb
3. Have you check manually that you see spindles in the signal? Could the amplitude of the signal be wrong, or noisy? You could try running the spindles detection with only 2 of the three thresholds to see if you get more detected spindles.

[qaqis95]

Thank you so much for your reply @raphaelvallat . I think this will be my final question.

The value of 620 was obtained through the 'sw = yasa.sw_detect(raw_eeg, sf, ch_names=[], hypno=hypno, include=(2, 3), coupling=True, coupling_params=dict(freq_sp=(12, 16), time=2, p=None))'.

Do you think this value (620) represents the total number of slow waves or the total number of slow waves coupled with sigma? I'm not quite sure about this, as I think the total number of slow waves may be obtained through 'sw = yasa.sw_detect(raw_eeg, amp_ptp=(50, 400))'.

The number of So obtained from these two functions differs significantly, so I still believe that 620 represents the number of SO-sigma coupling.

I apologize for any foolish questions I may have asked.

[raphaelvallat]

620 represents the total number of slow-waves. It is expected that the number will change if you manually change the detection thresholds by using: amp_ptp=(50, 400)). You can get the total number of slow-waves with some significant coupling with the sigma band by doing (sw_results["ndPAC"] > 0).sum(). However this will only work with the default coupling_params=dict(p=0.05) and not p=None.

Hope this helps,
Thanks