Model Card for ECGDenoiser

Collection: NeuralLib: Deep Learning Models for Biosignals Processing

Description: GRU-based model for ECG noise removal. Model and results published in the paper 'Cleaning ECG with Deep Learning: A Denoiser Tested in Industrial Settings'

• Architecture: GRUseq2seq
• Model Name: ECGDenoiser
• Task: ecg denoising: removing MA, BW and EM noise
• Train Dataset: PTB-XL+MIT-BIH-Noise-Stress-Test-Database

Biosignal(s): ECG

Sampling frequency: 360

Benchmark Results

Validation Loss: 0.0000

Training Time: 0.00 seconds

FLOPs per timestep: 0

Number of trainable parameters: 26121

Hyperparameters

Parameter	Value
bidirectional	True
dropout	0
hid_dim	[64, 1]
learning_rate	0.005
model_name	ECGDenoiser
multi_label	False
n_features	1
n_layers	2
num_classes	NA
task	regression
fc_out_bool	False

Example

import NeuralLib.model_hub as mh

model_name = ECGDenoiser()

model = mh.ProductionModel(model_name=model_name)

signal = torch.rand(1, 100, 1) # Example input signal

predictions = model.predict(signal)

print(predictions)