HFODetector

HFODetector is Python package that that is capable of detecting HFOs with either a STE or an MNI detector. Detection speed is increased by using multiprocessing.

Bibtex

If you find our project is useful in your research, please cite:

Zhang, Y., Liu, L., Ding, Y., Chen, X., Monsoor, T., Daida, A., Oana, S., Hussain, S. A., Sankar, R., Fallah, A., Santana-Gomez, C., Engel, J., Staba, R. J., Speier, W., Zhang, J., Nariai, H., & Roychowdhury, V. (2024). PyHFO: lightweight deep learning-powered end-to-end high-frequency oscillations analysis application. Journal of neural engineering, 10.1088/1741-2552/ad4916. Advance online publication. https://doi.org/10.1088/1741-2552/ad4916

Installation

pip install HFODetector

---

Run time comparison (in minutes)

	Linux	Linux	Windows	Windows	OS X	OS X
	STE	MNI	STE	MNI	STE	MNI
RIPPLELAB	372.83	5647.12	-	-	-	-
pyHFO single-core	57.43	971.35	34.57	933.31	35.90	659.63
pyHFO multi-core	5.18	83.30	9.03	113.59	7.56	114.35

The testing data we are using is 19 patients 10 min data in Refining epileptogenic high-frequency oscillations using deep learning: A novel reverse engineering approach paper.

** Single-core: n-jobs =1 for all machines,

** Multi-core: n-jobs = 32 for Linux (AMD Ryzen Threadripper 2950X), n-jobs = 8 for Windows(Intel i9-13900K) and Mac machines(Apple M1 Pro).

Example usage

STE detector

To use the STE detector, import ste from HFODetector, then a detector can be initialized with the desired parameters by calling ste.STEDetector. To use it to detect HFOs from an .edf file, call the detect_edf method. The detect_edf method takes a path to an edf file as input and returns a tuple containing the channel names and start and end timestamps of detected HFOs. The following code snippet shows how to use the STE detector.

import numpy as np
import pandas as pd
from HFODetector import ste

if __name__ == "__main__":
    edf_path = "example.edf" #change this to your edf path
    detector = ste.STEDetector(sample_freq=2000, filter_freq=[80, 500], 
                rms_window=3*1e-3, min_window=6*1e-3, min_gap=10 * 1e-3, 
                epoch_len=600, min_osc=6, rms_thres=5, peak_thres=3,
                n_jobs=32, front_num=1)
    channel_names, start_end = detector.detect_edf(edf_path)
    # channel_names is a list that is the same length as the number of channels in the edf
    # start_end is a nested list with the same length as channel_names. start_end[i][j][0] and start_end[i][j][1] 
    # will give the start and end index respectively for the jth detected HFO in channel channel_names[i]
    channel_names = np.concatenate([[channel_names[i]]*len(start_end[i]) for i in range(len(channel_names))])
    start_end = [start_end[i] for i in range(len(start_end)) if len(start_end[i])>0]
    start_end = np.concatenate(start_end)
    HFO_ste_df = pd.DataFrame({"channel":channel_names,"start":start_end[:,0],"end":start_end[:,1]})

Which results a pandas dataframe HFO_ste_df a sample is displayed below:

This dataframe has the following 3 columns:

• channel : name of the channel corresponding to the detected HFO
• start : start timestamp of the detected HFO in milliseconds
• end : end timestamp of the detected HFO in milliseconds

MNI detector

To use the MNI detector, import mni from HFODetector, then a detector can be initialized with the desired parameters by calling mni.MNIDetector. To use it to detect HFOs from an .edf file, call the detect_edf method. The detect_edf method takes a path to an edf file as input and returns a tuple containing the channel names and start and end timestamps of detected HFOs. The following code snippet shows how to use the MNI detector.

import numpy as np
import pandas as pd
from HFODetector import mni

if __name__ == "__main__":
    edf_path = "example_edf.edf" #change this to your edf path
    sample_freq=2000 #change this to your sample frequency
    detector = mni.MNIDetector(sample_freq, filter_freq=[80, 500], 
                epoch_time=10, epo_CHF=60, per_CHF=95/100, 
                min_win=10*1e-3, min_gap=10*1e-3, thrd_perc=99.9999/100, 
                base_seg=125*1e-3, base_shift=0.5, base_thrd=0.67, base_min=5,
                n_jobs=32, front_num=1)
    channel_names, start_end = detector.detect_edf(edf_path)
    # channel_names is a list that is the same length as the number of channels in the edf
    # start_end is a nested list with the same length as channel_names. start_end[i][j][0] and start_end[i][j][1] 
    # will give the start and end index respectively for the jth detected HFO in channel channel_names[i]
    channel_names = np.concatenate([[channel_names[i]]*len(start_end[i]) for i in range(len(channel_names))])
    start_end = [start_end[i] for i in range(len(start_end)) if len(start_end[i])>0]
    start_end = np.concatenate(start_end)
    HFO_mni_df = pd.DataFrame({"channel":channel_names,"start":start_end[:,0],"end":start_end[:,1]})

Which results a pandas dataframe HFO_mni_df a sample is displayed below:

This dataframe has the following 3 columns:

• channel : name of the channel corresponding to the detected HFO
• start : start timestamp of the detected HFO in milliseconds
• end : end timestamp of the detected HFO in milliseconds

Contributors:

Department of Electrical and Computer Engineering, University of California, Los Angeles

• Xin Chen -- Main Author of MNI
• Hoyoung Chung -- Main Author of STE
• Lawrence Liu
• Qiujing Lu
• Yuanyi Ding
• Yipeng Zhang

Division of Pediatric Neurology, Department of Pediatrics, UCLA Mattel Children’s Hospital David Geffen School of Medicine

• Hiroki Nariai

Contacts:

Please create a github issue or email the following people for further information

• Lawrence Liu (lawrencerliu@g.ucla.edu)
• Yuanyi Ding (semiswiet@g.ucla.edu)