python - 在 python 中加载 MIT-BIH 正常窦性心律数据库

Question

我正在尝试在 python 中加载 MIT-BIH 正常窦性心律数据库 (nsrdb)。我检查了这个教程:

https://medium.com/@roszcz/machine-learning-for-medicine-qrs-detection-in-a-single-channel-ecg-signal-part-1-data-set-be36f70bbd38

但是它用于加载 MIT-BIH 心律失常数据库。数据集文件夹中的 mitdb.py 文件中的以下代码如下所示。如何调整 get_records() 函数以下载其他 phyonet 数据库以及 nsrdb ？

Physionet数据库链接: https://www.physionet.org/physiobank/database/nsrdb/

import os
import h5py
import wfdb as wf
import numpy as np
import pandas as pd
from glob import glob
from scipy import signal as ss
from utils import download as ud
from matplotlib import pyplot as plt

def get_records():
    """ Get paths for data in data/mit/ directory """
    # Download if doesn't exist
    if not os.path.isdir('data/mitdb'):
        print 'Downloading the mitdb ecg database, please wait'
        ud.download_mitdb()
        print 'Download finished'

    # There are 3 files for each record
    # *.atr is one of them
    paths = glob('data/mitdb/*.atr')

    # Get rid of the extension
    paths = [path[:-4] for path in paths]
    paths.sort()

    return paths

def good_types():
    """ Of annotations """
    # www.physionet.org/physiobank/annotations.shtml
    good = ['N', 'L', 'R', 'B', 'A',
            'a', 'J', 'S', 'V', 'r',
            'F', 'e', 'j', 'n', 'E',
            '/', 'f', 'Q', '?']

    return good

def beat_annotations(annotation):
    """ Get rid of non-beat markers """
    # Declare beat types
    good = good_types()
    ids = np.in1d(annotation.anntype, good)

    # We want to know only the positions
    beats = annotation.annsamp[ids]

    return beats

def convert_input(channel, annotation):
    """ Into output """
    # Remove non-beat annotations
    beats = beat_annotations(annotation)

    # Create dirac-comb signal
    dirac = np.zeros_like(channel)
    dirac[beats] = 1.0

    # Use hamming window as a bell-curve filter
    width = 36
    filter = ss.hamming(width)
    gauss = np.convolve(filter, dirac, mode = 'same')

    return dirac, gauss

def good_annotations():
    """ Const function with good annotations """
    # For now it seems those are most popular
    good_annotations = [1, 2, 3, 4,
                        5, 6, 7, 8,
                        9, 10, 11, 12,
                        13, 16, 31, 38]

    return good_annotations

def make_dataset(records, width, savepath):
    """ Inside an array """
    # Prepare containers
    signals, labels = [], []

    # Iterate files
    for path in records:
        print 'Processing file:', path
        record = wf.rdsamp(path)
        annotations = wf.rdann(path, 'atr')

        # Extract pure signals
        data = record.p_signals

        # Convert each channel into labeled fragments
        signal, label = convert_data(data, annotations, width)

        # Cumulate
        signals.append(signal)
        labels.append(label)

    # Convert to one huge numpy.array
    signals = np.vstack(signals)
    labels = np.vstack(labels)

    # Write to disk
    np.save(savepath, {'signals' : signals,
                       'labels'  : labels })

def convert_data(data, annotations, width):
    """ Into a batch """
    # Prepare containers
    signals, labels = [], []

    # Convert both channels
    for it in range(2):
        channel = data[:, it]
        dirac, gauss = convert_input(channel,
                                     annotations)
        # Merge labels
        label = np.vstack([dirac, gauss])

        # Prepare the moving window
        sta = 0
        end = width
        stride = width
        while end <= len(channel):
            # Chop out the fragments
            s_frag = channel[sta : end]
            l_frag = label[:, sta : end]

            # Cumulate
            signals.append(s_frag)
            labels.append(l_frag)

            # Go forth
            sta += stride
            end += stride

    # Turn into arrays
    signals = np.array(signals)
    labels = np.array(labels)

    return signals, labels

def create_datasets():
    """ Training, validation, test """
    # Prepare paths
    records = get_records()

    # Shuffle up determinitically
    np.random.seed(666)
    np.random.shuffle(records)

    # Define the data
    width = 200

    # Make training
    make_dataset(records[:30], width, 'data/training')

    # ... validation ...
    make_dataset(records[30 : 39], width, 'data/validation')

    # ... and test
    make_dataset(records[39 : 48], width, 'data/test')

Answer 1

如果有人仍然想知道:有一个 python 包装器用于加载 WFDB Software Project用于从 Physionet 加载波形数据(从本地和远程加载)。

以下是如何使用它的示例:https://github.com/MIT-LCP/wfdb-python/blob/master/demo.ipynb

要加载示例，您可以在安装 wfdb 后运行以下行:

import wfdb
record = wfdb.rdrecord('16265', pb_dir='nsrdb/')
wfdb.plot_wfdb(record=record, title='Record 16265 from Physionet NSRDB') 
print(record.__dict__)

用于加载整个数据库使用

wfdb.dl_database('nsrdb', 'data/nsrdb')