forked from vitaldb/pyvital
-
Notifications
You must be signed in to change notification settings - Fork 0
/
ecg_annotator.py
331 lines (276 loc) · 11.2 KB
/
ecg_annotator.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
import numpy as np
import math
import arr
import copy
import pywt
cfg = {
'name': 'ECG - Annotator',
'group': 'Medical algorithms',
'desc': 'ECG annotator based on YC Chesnokov\'s implementation',
'reference': 'YC Chesnokov, D Nerukh, RC Glen, Individually Adaptable Automatic QT Detector',
'overlap': 3, # 2 sec overlap for HR=30
'interval': 30,
'inputs': [{'name': 'ecg', 'type': 'wav'}],
'outputs': [{'name': 'ann', 'type': 'str', 'unit': ''}],
'license': 'GPL'
}
def minimax(data):
return np.std(data) * (0.3936 + 0.1829 * math.log(len(data)))
def denoise(data, wsize):
# hard minmax denoise
for i in range(0, len(data), wsize):
iend = min(len(data), i + wsize)
th = minimax(data[i: iend])
for j in range(i, iend):
if abs(data[j]) <= th:
data[j] = 0
def cwt(data, srate, wname, freq):
scale = 0.16 * srate / freq # for gaus1
sig = pywt.cwt(data, [scale], wname)[0].flatten()
return sig
def qmf(w):
ret = []
for i in range(len(w)):
if i % 2 == 1:
ret.append(-w[len(w)-1-i])
else:
ret.append(w[len(w)-1-i])
return ret
def orthfilt(w):
lor = w / np.linalg.norm(w)
lod = lor[::-1]
hir = qmf(lor)
hid = hir[::-1]
return [lod, hid, lor, hir]
def run(inp, opt, cfg):
data = arr.interp_undefined(inp['ecg']['vals'])
srate = inp['ecg']['srate']
min_hr = 40 # min bpm
max_hr = 200 # max bpm
min_qrs = 0.04 # min qslist duration
max_qrs = 0.2 # max qslist duration
min_umv = 0.2 # min UmV of R,S peaks
min_pq = 0.07 # min PQ duration
max_pq = 0.20 # max PQ duration
min_qt = 0.21 # min QT duration
max_qt = 0.48 # max QT duration
pfreq = 9.0 # cwt Hz for pidx wave
tfreq = 2.5 # cwt Hz for tidx wave
min_sq = (60.0 / max_hr) - max_qrs # from s to next q
if min_sq * srate <= 0:
min_sq = 0.1
max_hr = int(60.0 / (max_qrs + min_sq))
# denoised ecg
depth = int(math.ceil(np.log2(srate / 0.8))) - 1
ad = pywt.wavedec(data, 'db2', level=depth)
ad[0].fill(0) # low frequency approx -> 0
ecg_denoised = pywt.waverec(ad, 'db2')
# interpolation filter
inter1 = pywt.Wavelet('inter1', filter_bank=orthfilt([0.25, 0.5, 0.25]))
# qrs augmented ecg
sig = cwt(data, srate, 'gaus1', 13) # 13 Hz gaus convolution
depth = int(math.ceil(np.log2(srate / 23))) - 2
ad = pywt.wavedec(sig, inter1, level=depth)
for level in range(depth): # remove [0-30Hz]
wsize = int(2 * srate / (2 ** (level+1))) # 2 sec window
denoise(ad[depth-level], wsize) # Remove less than 30 hz from all detail
ad[0].fill(0) # most lowest frequency approx -> 0
ecg_qrs = pywt.waverec(ad, inter1)
# start parsing
qslist = [] # qrs list [startqrs, endqrs, startqrs, endqrs, ...]
vpclist = [] # abnormal beat
# save greater than 0 after min_sq
prev_zero = 0
ipos = 0
while ipos < len(ecg_qrs) - int(max_qrs * srate):
if ecg_qrs[ipos] == 0:
prev_zero += 1
else:
if prev_zero > min_sq * srate:
iend = ipos + int(max_qrs * srate) # find the position of the end of the current qrs
while iend > ipos:
if ecg_qrs[iend] != 0:
break
iend -= 1
# Check if it is the minimum length or if there is a pause
if ipos + min_qrs * srate > iend or np.any(ecg_qrs[iend + 1:iend + 1 + int(min_sq * srate)]):
vpclist.append(ipos) # push vpc
else:
qslist.append(ipos)
qslist.append(iend)
ipos = iend
prev_zero = 0
ipos += 1
# qlist = [qslist[i] for i in range(0, len(qslist), 2)]
complist = []
for n in range(int(len(qslist) / 2)):
start_qrs = qslist[n * 2]
end_qrs = qslist[n * 2 + 1]
qidx = -1
ridx = arr.max_idx(ecg_denoised, start_qrs, end_qrs)
if ecg_denoised[ridx] < min_umv:
ridx = -1
sidx = arr.min_idx(ecg_denoised, start_qrs, end_qrs)
if -ecg_denoised[sidx] < min_umv:
sidx = -1
# ridxpeak > 0mV sidxpeak < 0mV
if ridx != -1 and sidx != -1:
if sidx < ridx: # check for sidx
if ecg_denoised[ridx] > -ecg_denoised[sidx]:
qidx = sidx
sidx = arr.min_idx(ecg_denoised, ridx, end_qrs + 1)
if sidx == ridx or sidx == end_qrs or abs(ecg_denoised[end_qrs] - ecg_denoised[sidx]) < 0.05:
sidx = -1
else: # check for qidx
qidx = arr.min_idx(ecg_denoised, start_qrs, ridx + 1)
if qidx == ridx or qidx == start_qrs or abs(ecg_denoised[start_qrs] - ecg_denoised[qidx]) < 0.05:
qidx = -1
elif sidx != -1: # only sidx --> Find small r if only sidx detected in rsidx large tidx lead
ridx = arr.max_idx(ecg_denoised, start_qrs, sidx + 1)
if ridx == sidx or ridx == start_qrs or abs(ecg_denoised[start_qrs] - ecg_denoised[ridx]) < 0.05:
ridx = -1
elif ridx != -1: # only ridx --> Find small q,s
qidx = arr.min_idx(ecg_denoised, start_qrs, ridx + 1)
if qidx == ridx or qidx == start_qrs or abs(ecg_denoised[start_qrs] - ecg_denoised[qidx]) < 0.05:
qidx = -1
sidx = arr.min_idx(ecg_denoised, ridx, end_qrs + 1)
if sidx == ridx or sidx == end_qrs or abs(ecg_denoised[end_qrs] - ecg_denoised[sidx]) < 0.05:
sidx = -1
else:
vpclist.append(start_qrs)
continue
o = {'q': qslist[n*2], 's': qslist[n*2+1]} # always exists
if qidx != -1:
o['q'] = qidx
if ridx != -1:
o['r'] = ridx
if sidx != -1:
o['s'] = sidx
complist.append(o)
# for each QRS --> find tidx and pidx wave
for n in range(len(complist) - 1):
pree = complist[n]['q']
nows = complist[n]['s']
nowe = complist[n+1]['q']
size = nowe - nows # s-q interval
size = int(min(size, srate * max_qt - (nows - pree)))
rr = (nowe - pree) / srate
if (60.0 / rr < min_hr) or (60.0 / rr > max_hr - 20):
continue
# all are in this
block = [data[nows + i] for i in range(size)]
ecg_qrs = cwt(block, srate, 'gaus1', tfreq)
tidx1 = arr.min_idx(ecg_qrs) + nows
tidx2 = arr.max_idx(ecg_qrs) + nows
if tidx1 > tidx2:
tidx1, tidx2 = tidx2, tidx1
# additional constraints on [tidx1 tidx tidx2] duration, symmetry, QT interval
ist = False
if ecg_qrs[tidx1-nows] < 0 < ecg_qrs[tidx2-nows]:
ist = True
elif ecg_qrs[tidx1-nows] > 0 > ecg_qrs[tidx2-nows]:
ist = True
if ist:
if (tidx2 - tidx1) >= 0.09 * srate: # and (tidx2-tidx1)<=0.24 * srate) #check for tidx wave duration
ist = True # QT interval = .4 * sqrt(RR)
if min_qt * srate <= (tidx2 - pree) <= max_qt * srate:
ist = True
else:
ist = False
else:
ist = False
if ist:
tidx = 0 # zero crossing
sign = (ecg_qrs[tidx1-nows] >= 0)
for i in range(tidx1 - nows, tidx2 - nows):
if sign == (ecg_qrs[i] >= 0):
continue
tidx = i + nows
break
# check for tidx wave symetry
if tidx2 - tidx < tidx - tidx1:
ratio = (tidx2 - tidx) / (tidx - tidx1)
else:
ratio = (tidx - tidx1) / (tidx2 - tidx)
if ratio < 0.4:
ist = False
if ist:
tmin = arr.min_idx(data, tidx1, tidx2)
tmax = arr.max_idx(data, tidx1, tidx2)
# find the most nearest values from 0-cross, tmin, tmax
tidx = arr.find_nearest((tidx, tmin, tmax), (tidx2 + tidx1) / 2)
complist[n]['(t'] = tidx1
complist[n]['t'] = tidx
complist[n]['t)'] = tidx2
# search for P-WAVE
size = nowe - nows # s-q interval
size = int(min(size, srate * max_pq))
if ist:
if tidx2 > nowe - size - int(0.04 * srate): # isp wnd far from Twave at least on 0.04 sec
size -= tidx2 - (nowe - size - int(0.04 * srate))
nskip = (nowe - nows) - size
if size <= 0.03 * srate:
continue # impresize QRS begin detection
block = [data[nows + nskip + i] for i in range(size)]
ecg_qrs = cwt(block, srate, 'gaus1', pfreq)
p1 = arr.min_idx(ecg_qrs) + nows + nskip
p2 = arr.max_idx(ecg_qrs) + nows + nskip
if p1 > p2:
p1, p2 = p2, p1
# additional constraints on [p1 pidx p2] duration, symmetry, PQ interval
isp = False
if ecg_qrs[p1-nows-nskip] < 0 < ecg_qrs[p2-nows-nskip]:
isp = True
elif ecg_qrs[p1-nows-nskip] > 0 > ecg_qrs[p2-nows-nskip]:
isp = True
if isp:
if 0.03 * srate <= (p2 - p1) <= 0.15 * srate: # check for pidx wave duration 9Hz0.03 5Hz0.05
isp = (min_pq * srate <= (nowe - p1) <= max_pq * srate) # PQ interval = [0.07 - 0.12,0.20]
else:
isp = False
if not isp:
continue
pidx = 0 # zero crossing
sign = (ecg_qrs[p1-nows-nskip] >= 0)
for i in range(p1 - nows - nskip, p2 - nows - nskip):
if sign == (ecg_qrs[i] >= 0):
continue
pidx = i + nows + nskip
break
# check for pidx wave symetry
if p2 - pidx < pidx - p1:
ratio = (p2 - pidx) / (pidx - p1)
else:
ratio = (pidx - p1) / (p2 - pidx)
if ratio < 0.4:
isp = False # not a p wave
if isp:
complist[n]['(p'] = p1
complist[n]['p'] = pidx
complist[n]['p)'] = p2
# add annotation
ret_ann = []
for n in range(len(complist)):
for k, v in complist[n].items():
if k == 'q' and abs(ecg_denoised[v]) > 0.5:
k = 'Q'
elif k == 'r' and abs(ecg_denoised[v]) > 0.5:
k = 'R'
elif k == 's' and abs(ecg_denoised[v]) > 0.5:
k = 'S'
elif k == '(t':
k = '(T'
elif k == 't':
k = 'T'
elif k == 't)':
k = 'T)'
elif k == '(p':
k = '(P'
elif k == 'p':
k = 'P'
elif k == 'p)':
k = 'P)'
ret_ann.append({"dt": v / srate, "val": k})
for n in range(len(vpclist)):
ret_ann.append({"dt": vpclist[n] / srate, "val": 'A'})
return [ret_ann]