BPD Final Presentation

Christopher Kok Yong En
Supervisor: Dr U. Rajendra Acharya

Project Title

Automated Classification of Arrhythmic Heartbeats using Deep Learning Technique.

Arrhythmia

Classifications

  • Non-Ectopic (N)
  • Supraventricular Ectopic (S)
  • Ventricular Ectopic (V)
  • Fusion (F)
  • Unknown (Q)


Beat classification according to the Association for the Advancement of Medical Instrumentation (AAMI)

WHO Statistics

33.5 million people worldwide -- or .5% of the world's population -- have the condition [Atrial Fibrillation].

Current Detection Methods

  • Manual Detection
    • Time Consuming
    • Prone to Human Error
  • Computer Aided Diagnosis
    • Can process multiple batches of data simultaneously
    • Requires much data to train accurately

Computer Aided Diagnosis

Machine Learning

  1. Detect Features
  2. Extract Features
  3. Model Compilation & Training
  4. Validation
Taken from: https://machinelearningblogcom.files.wordpress.com/2017/11/bildschirmfoto-2017-11-09-um-16-36-191.png

Deep Learning

Taken from: https://cdn-images-1.medium.com/max/1600/1*dnvGC-PORSoCo7VXT3PV_A.png

Deep Learning (Cont.)

Methodology

Data Used

Physionet MITDB Arrhythmia Database

Beat Classifications

Beat Classifications Count

Beat Type Beat Quantity
Non-Ectopic (N) 90, 631
Supraventricular Ectopic (S) 2, 781
Ventricular Ectopic (V) 7, 256
Fusion (F) 803
Unknown (Q) 1, 015

Data Augmentation and Synthesis

Convolutional Neural Network (CNN)

AlexNet CNN Architecture (Pedraza et al., 2017)

Definition

Taken from: https://cdn-images-1.medium.com/max/1600/1*7S266Kq-UCExS25iX_I_AQ.png

The output value of layer with input size \((N, C_{in}, L)\) and output \((N, C_{out}, L_{out})\) can be described as $$out(N_{i}, C_{out_{j}}) = bias(C_{out_{j}}) + \sum_{k=0}^{C_{in}-1}weight(C_{out_{j}}, k) \star input(N_{i}, k)$$
where \(\star\) is the cross-correlation operator, \(N\) is the batch size, \(C\) denotes a number of channels, \(L\) is a length of signal sequence, and $$L_{out}=\left \lfloor \frac{L_{in}+2\times \text{padding}-\text{dilation}\times(\text{kernel_size}-1)}{\text{stride}} +1 \right \rfloor$$

Implemented with open sourced library PyTorch (Adam et al., 2017)

Temporal Convolutional Network (TCN)

Visualization of a stack of dilated causal convolutional layers (Wavenet, 2016)

Model Architecture

TCN Block
Complete Architecture

Cyclic Learning Rates

Warm Restarts

Taken from: https://cdn-images-1.medium.com/max/2000/1*T5WWecP_EaQWk1yDX15h_w.png

Cosine Annealing

1-Cycle Policy

Results

SGD-WR Model

Confusion Matrix

Predicted acc (%) ppv (%) sen (%) spec (%)
N V
 S Q F
Original N 20098 3018 12568 2746 6861 79.91 99.17 44.38 87.74
V 66 44644 142 40 399 99.36 93.50 98.57 99.59
S 101 60 45106 22 2 99.82 78.02 99.59 99.88
Q 0 0 0 45291 0 100.00 94.16 100.00 100.00
F 1 26 0 0 45262 99.97 86.17 99.94 99.98
Average 95.81 90.20 88.50 97.44
acc = accuracy, ppv = positive predictive value, sen = sensitivity, spec = specificity

Metrics

1-Cycle AdamW Model

Confusion Matrix

Predicted acc (%) ppv (%) sen (%) spec (%)
N V
 S Q F
Original N 23920 2742 11072 2892 4665 82.59 98.85 52.81 89.32
V 49 44230 224 103 685 98.96 91.15 97.66 99.34
S 166 1190 44789 95 51 98.54 79.70 96.76 99.06
Q 19 64 58 45063 87 99.78 93.28 99.50 99.86
F 44 301 51 157 44738 99.46 89.07 98.78 99.65
Average 95.87 90.41 89.10 97.44
acc = accuracy, ppv = positive predictive value, sen = sensitivity, spec = specificity

Metrics

References

  • Adam, P., Soumith, C., Gregory, C., Edward, Y., Zachary, D., Zeming, L., … Adam, L. (2017). Automatic differentiation in pytorch. In Proceedings of Neural Information Processing Systems.
  • Pedraza, A., Gallego, J., Lopez, S., Gonzalez, L., Laurinavicius, A., & Bueno, G. (2017). Glomerulus Classification with Convolutional Neural Networks. In Communications in Computer and Information Science (pp. 839–849). Springer International Publishing. https://doi.org/10.1007/978-3-319-60964-5_73
  • van den Oord, A., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., … Kavukcuoglu, K. (2016). WaveNet: A Generative Model for Raw Audio. Retrieved from http://arxiv.org/abs/1609.03499

Q&A