-
Notifications
You must be signed in to change notification settings - Fork 48
/
ecg_annotation.py
324 lines (285 loc) · 10.8 KB
/
ecg_annotation.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
#import matplotlib with pdf as backend
import matplotlib
matplotlib.use('PDF')
import matplotlib.pyplot as plt
from matplotlib.backends.backend_pdf import PdfPages
import wfdb
import os
import numpy as np
import math
import sys
import scipy.stats as st
import glob, os
from os.path import basename
import tensorflow as tf
from keras.layers import Dense,Activation,Dropout
from keras.layers import LSTM,Bidirectional #could try TimeDistributed(Dense(...))
from keras.models import Sequential, load_model
from keras import optimizers,regularizers
from keras.layers.normalization import BatchNormalization
import keras.backend.tensorflow_backend as KTF
np.random.seed(0)
# functions
def get_ecg_data(datfile):
## convert .dat/q1c to numpy arrays
recordname=os.path.basename(datfile).split(".dat")[0]
recordpath=os.path.dirname(datfile)
cwd=os.getcwd()
os.chdir(recordpath) ## somehow it only works if you chdir.
annotator='q1c'
annotation = wfdb.rdann(recordname, extension=annotator, sampfrom=0,sampto = None, pbdir=None)
Lstannot=list(zip(annotation.sample,annotation.symbol,annotation.aux_note))
FirstLstannot=min( i[0] for i in Lstannot)
LastLstannot=max( i[0] for i in Lstannot)-1
print("first-last annotation:", FirstLstannot,LastLstannot)
record = wfdb.rdsamp(recordname, sampfrom=FirstLstannot,sampto = LastLstannot) #wfdb.showanncodes()
annotation = wfdb.rdann(recordname, annotator, sampfrom=FirstLstannot,sampto = LastLstannot) ## get annotation between first and last.
annotation2 = wfdb.Annotation(recordname='sel32', extension='niek', sample=(annotation.sample-FirstLstannot), symbol = annotation.symbol, aux_note=annotation.aux_note)
Vctrecord=np.transpose(record.p_signals)
VctAnnotationHot=np.zeros( (6,len(Vctrecord[1])), dtype=np.int)
VctAnnotationHot[5]=1 ## inverse of the others
#print("ecg, 2 lead of shape" , Vctrecord.shape)
#print("VctAnnotationHot of shape" , VctAnnotationHot.shape)
#print('plotting extracted signal with annotation')
#wfdb.plotrec(record, annotation=annotation2, title='Record 100 from MIT-BIH Arrhythmia Database', timeunits = 'seconds')
VctAnnotations=list(zip(annotation2.sample,annotation2.symbol)) ## zip coordinates + annotations (N),(t) etc)
#print(VctAnnotations)
for i in range(len(VctAnnotations)):
#print(VctAnnotations[i]) # Print to display annotations of an ecg
try:
if VctAnnotations[i][1]=="p":
if VctAnnotations[i-1][1]=="(":
pstart=VctAnnotations[i-1][0]
if VctAnnotations[i+1][1]==")":
pend=VctAnnotations[i+1][0]
if VctAnnotations[i+3][1]=="N":
rpos=VctAnnotations[i+3][0]
if VctAnnotations[i+2][1]=="(":
qpos=VctAnnotations[i+2][0]
if VctAnnotations[i+4][1]==")":
spos=VctAnnotations[i+4][0]
for ii in range(0,8): ## search for t (sometimes the "(" for the t is missing )
if VctAnnotations[i+ii][1]=="t":
tpos=VctAnnotations[i+ii][0]
if VctAnnotations[i+ii+1][1]==")":
tendpos=VctAnnotations[i+ii+1][0]
# #print(ppos,qpos,rpos,spos,tendpos)
VctAnnotationHot[0][pstart:pend]=1 #P segment
VctAnnotationHot[1][pend:qpos]=1 #part "nothing" between P and Q, previously left unnanotated, but categorical probably can't deal with that
VctAnnotationHot[2][qpos:rpos]=1 #QR
VctAnnotationHot[3][rpos:spos]=1 #RS
VctAnnotationHot[4][spos:tendpos]=1 #ST (from end of S to end of T)
VctAnnotationHot[5][pstart:tendpos]=0 #tendpos:pstart becomes 1, because it is inverted above
except IndexError:
pass
Vctrecord=np.transpose(Vctrecord) # transpose to (timesteps,feat)
VctAnnotationHot=np.transpose(VctAnnotationHot)
os.chdir(cwd)
return Vctrecord, VctAnnotationHot
def splitseq(x,n,o):
#split seq; should be optimized so that remove_seq_gaps is not needed.
upper=math.ceil( x.shape[0] / n) *n
print("splitting on",n,"with overlap of ",o, "total datapoints:",x.shape[0],"; upper:",upper)
for i in range(0,upper,n):
#print(i)
if i==0:
padded=np.zeros( ( o+n+o,x.shape[1]) ) ## pad with 0's on init
padded[o:,:x.shape[1]] = x[i:i+n+o,:]
xpart=padded
else:
xpart=x[i-o:i+n+o,:]
if xpart.shape[0]<i:
padded=np.zeros( (o+n+o,xpart.shape[1]) ) ## pad with 0's on end of seq
padded[:xpart.shape[0],:xpart.shape[1]] = xpart
xpart=padded
xpart=np.expand_dims(xpart,0)## add one dimension; so that you get shape (samples,timesteps,features)
try:
xx=np.vstack( (xx,xpart) )
except UnboundLocalError: ## on init
xx=xpart
print("output: ",xx.shape)
return(xx)
def remove_seq_gaps(x,y):
#remove parts that are not annotated <- not ideal, but quickest for now.
window=150
c=0
cutout=[]
include=[]
print("filterering.")
print("before shape x,y",x.shape,y.shape)
for i in range(y.shape[0]):
c=c+1
if c<window :
include.append(i)
if sum(y[i,0:5])>0:
c=0
if c >= window:
#print ('filtering')
pass
x,y=x[include,:],y[include,:]
print(" after shape x,y",x.shape,y.shape)
return(x,y)
def normalizesignal(x):
x=st.zscore(x, ddof=0)
return x
def normalizesignal_array(x):
for i in range(x.shape[0]):
x[i]=st.zscore(x[i], axis=0, ddof=0)
return x
def plotecg(x,y,begin,end):
#helper to plot ecg
plt.figure(1,figsize=(11.69,8.27))
plt.subplot(211)
plt.plot(x[begin:end,0])
plt.subplot(211)
plt.plot(y[begin:end,0])
plt.subplot(211)
plt.plot(y[begin:end,1])
plt.subplot(211)
plt.plot(y[begin:end,2])
plt.subplot(211)
plt.plot(y[begin:end,3])
plt.subplot(211)
plt.plot(y[begin:end,4])
plt.subplot(211)
plt.plot(y[begin:end,5])
plt.subplot(212)
plt.plot(x[begin:end,1])
plt.show()
def plotecg_validation(x,y_true,y_pred,begin,end):
#helper to plot ecg
plt.figure(1,figsize=(11.69,8.27))
plt.subplot(211)
plt.plot(x[begin:end,0])
plt.subplot(211)
plt.plot(y_pred[begin:end,0])
plt.subplot(211)
plt.plot(y_pred[begin:end,1])
plt.subplot(211)
plt.plot(y_pred[begin:end,2])
plt.subplot(211)
plt.plot(y_pred[begin:end,3])
plt.subplot(211)
plt.plot(y_pred[begin:end,4])
plt.subplot(211)
plt.plot(y_pred[begin:end,5])
plt.subplot(212)
plt.plot(x[begin:end,1])
plt.subplot(212)
plt.plot(y_true[begin:end,0])
plt.subplot(212)
plt.plot(y_true[begin:end,1])
plt.subplot(212)
plt.plot(y_true[begin:end,2])
plt.subplot(212)
plt.plot(y_true[begin:end,3])
plt.subplot(212)
plt.plot(y_true[begin:end,4])
plt.subplot(212)
plt.plot(y_true[begin:end,5])
def LoaddDatFiles(datfiles):
for datfile in datfiles:
print(datfile)
if basename(datfile).split(".",1)[0] in exclude:
continue
qf=os.path.splitext(datfile)[0]+'.q1c'
if os.path.isfile(qf):
#print("yes",qf,datfile)
x,y=get_ecg_data(datfile)
x,y=remove_seq_gaps(x,y)
x,y=splitseq(x,1000,150),splitseq(y,1000,150) ## create equal sized numpy arrays of n size and overlap of o
x = normalizesignal_array(x)
## todo; add noise, shuffle leads etc. ?
try: ## concat
xx=np.vstack( (xx,x) )
yy=np.vstack( (yy,y) )
except NameError: ## if xx does not exist yet (on init)
xx = x
yy = y
return(xx,yy)
def unison_shuffled_copies(a, b):
assert len(a) == len(b)
p = np.random.permutation(len(a))
return a[p], b[p]
def get_session(gpu_fraction=0.8):
#allocate % of gpu memory.
num_threads = os.environ.get('OMP_NUM_THREADS')
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=gpu_fraction)
if num_threads:
return tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, intra_op_parallelism_threads=num_threads))
else:
return tf.Session(config=tf.ConfigProto(gpu_options=gpu_options))
def getmodel():
model = Sequential()
model.add(Dense(32,W_regularizer=regularizers.l2(l=0.01), input_shape=(seqlength, features)))
model.add(Bidirectional(LSTM(32, return_sequences=True)))#, input_shape=(seqlength, features)) ) ### bidirectional ---><---
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(Dense(64, activation='relu',W_regularizer=regularizers.l2(l=0.01)))
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(Dense(dimout, activation='softmax'))
adam = optimizers.adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy'])
print(model.summary())
return(model)
##################################################################
##################################################################
qtdbpath=sys.argv[1] ## first argument = qtdb database from physionet.
perct=0.81 #percentage training
percv=0.19 #percentage validation
exclude = set()
exclude.update(["sel35","sel36","sel37","sel50","sel102","sel104","sel221","sel232", "sel310"])# no P annotated:
##################################################################
# datfile=qtdbpath+"sel49.dat" ## single ECG to test if loading works.
# x,y=get_ecg_data(datfile)
# print(x.shape,y.shape)
# # for i in range(y.shape[0]): #Invert QT-label to actually represent QT. Does give overlapping labels
# # y[i][0] = 1 - y[i][0]
# plotecg(x,y,0,y.shape[0]) ## plot all
# x,y=remove_seq_gaps(x,y) ## remove 'annotation gaps'
# plotecg(x,y,0,y.shape[0]) ## plot all
# x,y=splitseq(x,750,150),splitseq(y,750,150) ## create equal sized numpy arrays of n size and overlap of o
# exit()
##################################################################
# load data
datfiles=glob.glob(qtdbpath+"*.dat")
xxt,yyt=LoaddDatFiles(datfiles[ :round(len(datfiles)*perct) ]) # training data.
xxt,yyt=unison_shuffled_copies(xxt,yyt) ### shuffle
xxv,yyv=LoaddDatFiles(datfiles[ -round(len(datfiles)*percv): ] ) ## validation data.
seqlength=xxt.shape[1]
features=xxt.shape[2]
dimout=yyt.shape[2]
print("xxv/validation shape: {}, Seqlength: {}, Features: {}".format(xxv.shape[0],seqlength,features))
# #plot validation ecgs
# with PdfPages('ecgs_xxv.pdf') as pdf:
# for i in range( xxv.shape[0] ):
# print (i)
# plotecg(xxv[i,:,:],yyv[i,:,:],0,yyv.shape[1])
# pdf.savefig()
# plt.close()
# call keras/tensorflow and build lstm model
KTF.set_session(get_session())
with tf.device('/cpu:0'): #switch to /cpu:0 to use cpu
if not os.path.isfile('model.h5'):
model = getmodel() # build model
model.fit(xxt, yyt, batch_size=32, epochs=100, verbose=1) # train the model
model.save('model.h5')
model = load_model('model.h5')
score, acc = model.evaluate(xxv, yyv, batch_size=4, verbose=1)
print('Test score: {} , Test accuracy: {}'.format(score, acc))
# predict
yy_predicted = model.predict(xxv)
# maximize probabilities of prediction.
for i in range(yyv.shape[0]):
b = np.zeros_like(yy_predicted[i,:,:])
b[np.arange(len(yy_predicted[i,:,:])), yy_predicted[i,:,:].argmax(1)] = 1
yy_predicted[i,:,:] = b
# plot:
with PdfPages('ecg.pdf') as pdf:
for i in range( xxv.shape[0] ):
print (i)
plotecg_validation(xxv[i,:,:],yy_predicted[i,:,:],yyv[i,:,:],0,yy_predicted.shape[1]) # top = predicted, bottom=true
pdf.savefig()
plt.close()
#plotecg(xv[1,:,:],yv[1,:,:],0,yv.shape[1]) ## plot first seq