Here, I experimented with annotating peaks of the ECG using a recurrent neural network in tensorflow's Keras.
In the beginning I struggled a bit to get the input/output right, which had to do with the way I tried to format ECG-peaks (as a sparse vector containing peaks (1) vs no peaks (0)). Aproaching it as a semantic segmentation problem (e.g. Seq2Seq) solved it for me.

It seems to work well on the QT database of physionet, but there are some cases that it has never seen where it fails; I haven't played with augmenting the ecgs yet.

[5oct 2019] Since posting this 3 years ago, I noticed several publications using the (exact) same principle:

- Matlab: https://www.mathworks.com/help/signal/examples/waveform-segmentation-using-deep-learning.html
- IEEE: https://ieeexplore.ieee.org/document/8333406

```
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'])
```

- Download ECG data using something like
`wget -r -l1 --no-parent https://physionet.org/physiobank/database/qtdb/`

- Run, with as the first argument the directory where the ECG data is stored; or set
`qtdbpath`

.

A 2 lead ECG, the colors indicate P-Pend(yellow),Pend-Q(green), Q-R(red),R-S(purple),S-Tend(brown),Tend-P(pink). Training took about an hour on 6 cores.

- Colors at the top show true annotations
- Colors at the bottom show predicted annotations

I haven't got a list of all dependencies; but use this:

- wfdb 1.3.4 ( not the newest >2.0, thanks BrettMontague); pip install wfdb==1.3.4

Get A Weekly Email With Trending Projects For These Topics

No Spam. Unsubscribe easily at any time.

python (55,543)

deep-learning (4,081)

tensorflow (2,186)

tutorial (988)

segmentation (282)

lstm (269)

example (249)

time-series (249)

annotations (115)

keras-tutorials (19)

ecg (17)