mitbih-ptbdb-ecg-heartbeat-categorization

We used these datasets to train a CNN and an MLP.

Datasets

MITBIH

Testing Set Count	Training Set Count	Category
18118	72471	Category - N
556	2223	Category - S
1448	5788	Category - V
162	641	Category - F
1608	6431	Category - Q

Samples Count: 109446

Categories Count: 5

Sampling Frequency: 125Hz

Data Source: Physionet's MIT-BIH Arrhythmia Dataset

Categories:

• N : Non-ecotic beats (normal beat)
• S : Supraventricular ectopic beats
• V : Ventricular ectopic beats
• F : Fusion Beats
• Q : Unknown Beats

PTBDB

4046	Normal
10506	Abnormal

Samples Count: 14552

Categories Count: 2

Sampling Frequency: 125Hz

Data Source: Physionet's PTB Diagnostic Database

Data Balancing

We used Fourier method to generate more signals so all the categories are of the same size.

Neural Networks Architectures

CNN

Layer (type)	Output Shape	#Param	Activation Function
Conv1D-64/7 (Conv1D)	(None, 181, 64)	512	ReLU
SD1D-0.4 (SpatialDropout1D)	(None, 181, 64)	0	-
BN-1 (BatchNormalization)	(None, 181, 64)	256	-
MP1D-7 (MaxPooling1D)	(None, 25, 64)	0	-
Conv1D-32/5 (Conv1D)	(None, 21, 32)	10272	ReLU
BN-2 (BatchNormalization)	(None, 21, 32)	128	-
MP1D-5 (MaxPooling1D)	(None, 4, 32)	0	-
GMP1D (GlobalMaxPooling1D)	(None, 32)	0	-
Output (Dense)	(None, 5)	165	Softmax / Sigmoid
Total params: 11,333
Trainable params: 11,141
Non-trainable params: 192

MLP

Layer (type)	Output Shape	#Param	Activation Function
Dense-64 (Dense)	(None, 64)	12032	ReLU
Dropout-0.25 (Dropout)	(None, 64)	0	-
Dense-32 (Dense)	(None, 32)	2080	ReLU
Dense-16 (Dense)	(None, 16)	528	ReLU
Output (Dense)	(None, 5)	85	Softmax
Total params: 14,725
Trainable params: 14,725
Non-trainable params: 0

Results

CNN-MITBIH

	Precision	Recall	f1-score	Support
Category-N	0.98	0.98	0.98	18118
Category-S	0.73	0.60	0.65	556
Category-V	0.94	0.91	0.93	1448
Category-F	0.49	0.85	0.62	162
Category-Q	0.96	0.99	0.97	1608
Accuracy	0.9658322930335999			21892
Loss	0.13102902472019196			21892
Macro Avg	0.82	0.86	0.83	21892
Weighted Avg	0.97	0.97	0.97	21892

CNN-PTBDB

	Precision	Recall	f1-score	Support
Normal	0.97	0.99	0.98	818
Abnormal	1.00	0.99	0.99	2093
Accuracy	0.9900377988815308			2911
Loss	0.0375625379383564			2911
Macro Avg	0.99	0.99	0.99	2911
Weighted Avg	0.99	0.99	0.99	2911

MLP-MITBIH

	Precision	Recall	f1-score	Support
Category-N	0.97	0.99	0.98	18118
Category-S	0.76	0.60	0.67	556
Category-V	0.93	0.84	0.88	1448
Category-F	0.72	0.64	0.68	162
Category-Q	0.98	0.92	0.95	1608
Accuracy	0.9611730575561523			21892
Loss	0.15909039974212646			21892
Macro Avg	0.87	0.80	0.83	21892
Weighted Avg	0.96	0.96	0.96	21892

MLP-PTBDB

	Precision	Recall	f1-score	Support
Normal	0.93	0.94	0.94	818
Abnormal	0.98	0.97	0.97	2093
Accuracy	0.9639298915863037			2911
Loss	0.12964512407779694			2911
Macro Avg	0.95	0.96	0.96	2911
Weighted Avg	0.96	0.96	0.96	2911

License

Unlicense