import os from collections import defaultdict import pyedflib import pyedflib.highlevel as hl import numpy as np import copy import shutil import bz2 import pickle import _pickle as cPickle import multiprocessing as mp # Pickle a file and then compress it into a file with extension def compressed_pickle(title, data): # with bz2.BZ2File(title + '.pbz2', 'w') as f: # cPickle.dump(data, f) pickle.dump(data, open(title, "wb")) # Process metadata def process_metadata(summary, filename): f = open(summary, "r") metadata = {} lines = f.readlines() times = [] for i in range(len(lines)): line = lines[i].split() if len(line) == 3 and line[2] == filename: j = i + 1 processed = False while not processed: if lines[j].split()[0] == "Number": seizures = int(lines[j].split()[-1]) processed = True j = j + 1 # If file has seizures get start and end time if seizures > 0: j = i + 1 for s in range(seizures): # Save start and end time of each seizure processed = False while not processed: l = lines[j].split() # print(l) if l[0] == "Seizure" and "Start" in l: start = int(l[-2]) * 256 - 1 # Index of start time end = ( int(lines[j + 1].split()[-2]) * 256 - 1 ) # Index of end time processed = True j = j + 1 times.append((start, end)) metadata["seizures"] = seizures metadata["times"] = times return metadata # Keep some channels from a .edf and ignore the others def drop_channels(edf_source, edf_target=None, to_keep=None, to_drop=None): signals, signal_headers, header = hl.read_edf( edf_source, ch_nrs=to_keep, digital=False ) clean_file = {} for signal, header in zip(signals, signal_headers): channel = header.get("label") if channel in clean_file.keys(): channel = channel + "-2" clean_file[channel] = signal return clean_file # At first, it permuted the channels of a edf signal # Now, only keeps valid channels and compress+save into pkl def move_channels(clean_dict, channels, target): # Keep only valid channels keys_to_delete = [] for key in clean_dict: if key != "metadata" and key not in channels.keys(): keys_to_delete.append(key) for key in keys_to_delete: del clean_dict[key] # Get size of the numpy array size = 0 for item in clean_dict.keys(): if item != "metadata": size = len(clean_dict.get(item)) break for k in channels.keys(): if k not in clean_dict.keys(): clean_dict[k] = np.zeros(size, dtype=float) compressed_pickle(target + ".pkl", clean_dict) # Process edf files of a pacient from start number to end number def process_files(pacient, valid_channels, channels, start, end): for num in range(start, end + 1): to_keep = [] num = ("0" + str(num))[-2:] filename = "{path}/chb{p}/chb{p}_{n}.edf".format( path=signals_path, p=pacient, n=num ) # Check with (cleaned) reference file if we have to remove more channels try: signals, signal_headers, header = hl.read_edf(filename, digital=False) n = 0 for h in signal_headers: if h.get("label") in valid_channels: if n not in to_keep: to_keep.append(n) n = n + 1 except OSError: print("****************************************") print("WARNING - Do not worry") print("File", filename, "does not exist.\nProcessing next file.") print("****************************************") continue if len(to_keep) > 0: try: print( "Removing", len(signal_headers) - len(to_keep), "channels from file ", "chb{p}_{n}.edf".format(p=pacient, n=num), ) clean_dict = drop_channels( filename, edf_target="{path}/chb{p}/chb{p}_{n}.edf".format( path=clean_path, p=pacient, n=num ), to_keep=to_keep, ) print("Processing file ", filename) except AssertionError: print("****************************************") print("WARNING - Do not worry") print("File", filename, "does not exist.\nProcessing next file.") print("****************************************") continue metadata = process_metadata( "{path}/chb{p}/chb{p}-summary.txt".format(path=signals_path, p=pacient), "chb{p}_{n}.edf".format(p=pacient, n=num), ) metadata["channels"] = valid_channels clean_dict["metadata"] = metadata target = "{path}/chb{p}/chb{p}_{n}.edf".format( path=clean_path, p=pacient, n=num ) move_channels(clean_dict, channels, target) def start_process(pacient, num, start, end, sum_ind): # Summary file f = open( "{path}/chb{p}/chb{p}-summary.txt".format(path=signals_path, p=pacient), "r" ) channels = defaultdict(list) # Dict of channels and indices valid_channels = [] # Valid channels to_keep = [] # Indices of channels we want to keep channel_index = 1 # Index for each channel summary_index = 0 # Index to choose which channel reference take from summary file # Process summary file for line in f: line = line.split() if len(line) == 0: continue if line[0] == "Channels" and line[1] == "changed:": summary_index += 1 if ( line[0] == "Channel" and summary_index == sum_ind and (line[2] != "-" and line[2] != ".") ): # '-' means a void channel if ( line[2] in channels.keys() ): # In case of repeated channel just add '-2' to the label name = line[2] + "-2" else: name = line[2] # Add channel to dict and update lists channels[name].append(str(channel_index)) channel_index += 1 valid_channels.append(name) to_keep.append(int(line[1][:-1]) - 1) # for item in channels.items(): print(item) # Clean reference file filename = "{path}/chb{p}/chb{p}_{n}.edf".format( path=signals_path, p=pacient, n=num ) target = "{path}/chb{p}/chb{p}_{n}.edf".format(path=clean_path, p=pacient, n=num) if not os.path.exists("{path}/chb{p}".format(p=pacient, path=clean_path)): os.makedirs("{path}/chb{p}".format(p=pacient, path=clean_path)) clean_dict = drop_channels(filename, edf_target=target, to_keep=to_keep) # Process metadata : Number of seizures and start/end time metadata = process_metadata( "{path}/chb{p}/chb{p}-summary.txt".format(path=signals_path, p=pacient), "chb{p}_{n}.edf".format(p=pacient, n=num), ) metadata["channels"] = valid_channels clean_dict["metadata"] = metadata compressed_pickle(target + ".pkl", clean_dict) # Process the rest of the files to get same channels as reference file process_files(pacient, valid_channels, channels, start, end) # PARAMETERS signals_path = r"/data/datasets/chb-mit-scalp-eeg-database-1.0.0" # Path to the data main directory clean_path = r"/data/datasets/BigDownstream/chb-mit/processed" # Path where to store clean data if not os.path.exists(clean_path): os.makedirs(clean_path) # Clean pacients one by one manually with these parameters pacient = "04" num = "01" # Reference file summary_index = 0 # Index of channels summary reference start = 28 # Number of first file to process end = 28 # Number of last file to process # Start the process # start_process(pacient, num, start, end, summary_index) # FULL DATA PROCESS parameters = [ ("01", "01", 2, 46, 0), ("02", "01", 2, 35, 0), ("03", "01", 2, 38, 0), ("05", "01", 2, 39, 0), ("06", "01", 2, 24, 0), ("07", "01", 2, 19, 0), ("08", "02", 3, 29, 0), ("10", "01", 2, 89, 0), ("11", "01", 2, 99, 0), ("14", "01", 2, 42, 0), ("20", "01", 2, 68, 0), ("21", "01", 2, 33, 0), ("22", "01", 2, 77, 0), ("23", "06", 7, 20, 0), ("24", "01", 3, 21, 0), ("04", "07", 1, 43, 1), ("09", "02", 1, 19, 1), ("15", "02", 1, 63, 1), ("16", "01", 2, 19, 0), ("18", "02", 1, 36, 1), ("19", "02", 1, 30, 1), ] # parameters = [ # ("12", "") # ] with mp.Pool(mp.cpu_count()) as pool: res = pool.starmap(start_process, parameters)