ECG Deep Denoiser and Delineator

This project focuses on ECG denoising and delineation using deep learning techniques. The following steps will guide you through setting up the environment and running the training and evaluation scripts. Trained on LUDB and Evaluated on QTDB.

Architecture

Setup Instructions

1. Download Dataset for Windows: Execute the following command to download the required dataset:

.\download_dataset.cmd

2. Create Virtual Environment: Create a virtual environment to manage dependencies:

python -m venv .venv

3. Install Dependencies: Install the necessary Python packages using the provided requirements file:

pip install -r requirements.txt

Denoising

For detailed instructions of the denoising model, please refer to the Deep Filter repository: Deep Filter

Delineation

• Preprocessing: Preprocesses the delineation data locally for training purposes.

  This function performs preprocessing on the delineation data and returns the results in two formats:

  • .npz: A compressed file format that stores multiple NumPy arrays.
  • .pkl: A pickle file that serializes Python objects.

• Training: Follow the training notebook available on Kaggle: ECG Delineation 80Hz 256WinSize

• Evaluation on Unseen Data: Evaluate the model on unseen data using the following Kaggle notebook: QTDB Unseen Eval ECG Delineation 80Hz 256WinSize

Result

QTDB Unseen Overall

Classes	PPV	TPR	F1	TNR	NPV	ACC
BL	82.1	89	85.4	78.6	86.6	84
QRS	90	73.3	80.8	98.7	95.9	95.2
T	87.3	80.1	83.6	96.5	94.1	92.7
P	74.9	75.3	75.1	97	97.1	94.7
Overall Accuracy						83.3

QTDB Unseen Per Subject

Record	Lead	Pathology	Classes	PPV	TPR	F1	TNR	NPV	ACC
sel32	ECG1	Sudden Death (BIH)	BL	75.3	95.8	84.3	73.9	95.4	83.8
			QRS	96.6	55.8	70.7	99.6	91.3	91.8
			T	90.7	80.2	85.1	97.3	93.7	93.1
			P	92.7	77.5	84.4	99.2	97	96.6
			Overall Accuracy						82.6
sel32	ECG2	Sudden Death (BIH)	BL	75.4	77.4	76.4	66.9	69.2	72.8
			QRS	93.3	64.6	76.3	99	92.6	92.7
			T	51.4	55.2	53.2	89.7	91.1	84.1
			P	52.3	70.3	60	94.1	97.1	92
			Overall Accuracy						70.8
sel49	ECG1	Sudden Death (BIH)	BL	63.5	88.4	73.9	50.8	81.9	69.3
			QRS	77.8	67.3	72.2	97.5	95.9	94.1
			T	92.6	36.8	52.7	98.5	76.1	78.3
			P	60.9	79.8	69.1	96.4	98.6	95.3
			Overall Accuracy						68.5
sel49	ECG2	Sudden Death (BIH)	BL	72.4	94.7	82.1	61.1	91.4	78.5
			QRS	94.5	38.4	54.6	99.5	88.6	89
			T	85.3	70.5	77.2	96.1	91	89.8
			P	30.2	23.1	26.1	96.3	94.8	91.6
			Overall Accuracy						74.5
sel14046	ECG1	MIT-BIH Long-Term ECG	BL	85.7	88.8	87.2	82.9	86.5	86.1
			QRS	91.6	81.1	86	98.9	97.2	96.6
			T	90.7	85.5	88	97	95.1	94
			P	65.9	75	70.2	96.8	97.9	95.2
			Overall Accuracy						85.9
sel14046	ECG2	MIT-BIH Long-Term ECG	BL	95	78.7	86.1	93.5	73.9	84.5
			QRS	68.6	92.2	78.7	95.7	99.2	95.4
			T	70.9	89.8	79.3	92.2	97.7	91.8
			P	81.2	98.1	88.9	96.8	99.7	96.9
			Overall Accuracy						84.3
sel15814	ECG1	MIT-BIH Long-Term ECG	BL	74.8	81.7	78.1	81.3	86.7	81.4
			QRS	99	75.8	85.8	99.8	93.2	94.2
			T	92.2	85.4	88.6	97.3	94.6	94
			P	56	79	65.6	93.8	97.8	92.5
			Overall Accuracy						81.1
sel15814	ECG2	MIT-BIH Long-Term ECG	BL	69.3	95.1	80.2	68.1	94.9	79.7
			QRS	99	73.7	84.5	99.8	93.2	94.1
			T	94.7	79.6	86.5	98.4	93.2	93.5
			P	59.4	17.6	27.2	98.8	92.4	91.5
			Overall Accuracy						79.4