deepdelineator

The deepdelineator library can be used to detect characteristic points in arterial blood pressure waveforms using deep learning techniques to improve the accuracy of the detection. At the same time, regions with noise can be identified automatically.

Installation

pip install deepdelineator

Usage

pred_from_numpy()

Description:
This is the main function to obtain Fiducial Points (FiPs) from input signals.

---

Inputs

1. signal_list (list):
   A list containing multiple signals, each represented as a NumPy ndarray. Each ndarray must have the shape (samples,).

2. s_f (int):
   The sampling frequency of the signals. All signals in signal_list must share the same sampling frequency.

---

Return

detections (dict):
A dictionary where each key is the index of a signal (e.g., detections[idx]). Each value is a dictionary containing two main components:

1. resampled: Information about the signal analyzed by the model at the model's sampling frequency (rs_f = 125 Hz).
2. original: Information about the signal resampled to the provided sampling frequency (s_f).

---

Detection Structure

For each signal index (detections[idx]), the structure is as follows:

• resample: In terms of the model's sampling frequency (125 Hz).

resampled:

• signal: The resampled version of the signal analyzed at the model's sampling frequency (125 Hz).
• predictions: Class predictions made by the model.
• onset: Detected onset indices.
• peaks: Detected peak indices.
• dn: Detected DN (decay or downward movement) indices.
• beats_fips: List of completed beats, each represented as [N° beat, start, peak, dn, end].
• artifacts: A list of artifact indices.
• rs_f: The resampled frequency, fixed at 125 Hz.

original:

• signal: The analyzed signal resampled .
• predictions: Interpolated class predictions to match the given s_f.
• onset: Onset indices.
• peaks: Peak indices.
• dn: DN indices.
• beats_fips: List of completed beats (indices).
• artifacts: A list of artifact indices.
• s_f: The given s_f.

Note: original indices are in terms of the provided sampling frequency (s_f).

from deepdelineator.utils import load_delineator

delineator = load_delineator()
s_f = 500 # Sampling frecuency of the signals, adjust to your data
# Detection!
detections = model.pred_from_numpy(signal_list=list_of_abp_signals, s_f=s_f)

For more details, please refer to the examples folder.

Citing

If you consider this work useful, please cite the following paper:

A Delineator for Arterial Blood Pressure Waveform Analysis Based on a Deep Learning Technique

@INPROCEEDINGS{9630717,
  author={Aguirre, Nicolas and Grall-Maës, Edith and Cymberknop, Leandro J. and Armentano, Ricardo L.},
  booktitle={2021 43rd Annual International Conference of the IEEE Engineering in Medicine   Biology Society (EMBC)}, 
  title={A Delineator for Arterial Blood Pressure Waveform Analysis Based on a Deep Learning Technique}, 
  year={2021},
  volume={},
  number={},
  pages={56-59},
  doi={10.1109/EMBC46164.2021.9630717}}