search

inventec-ai-center / bp-benchmark

A Benchmark for Machine-Learning based Non-Invasive Blood Pressure Estimation using Photoplethysmogram
https://doi.org/10.1038/s41597-023-02020-6
MIT License
45 stars 17 forks source link

How about your work on MABP estimation? #7

Closed ms-keliu closed 5 months ago

ms-keliu commented 5 months ago

I've noticed "process_MABP.py" in your project to calculate mean ABP value during a cycle, which is not mentioned in your paper. Can you describe your work related to MABP? Thank you for your attention:)

sergiogvz commented 5 months ago

Thank you for your interest in our work. The file process_MABP.py aims to preprocess the ABP signal to create the ground-truth output for the PPG2IABP model [12], which is the average ABP pulse (MABP) with some semantic segmentation (systolic phase, Systolic peak to Dicrotic Notch, Dicrotic Notch to the end). If I don't remember wrongly, the original preprocessing code of Aguirre et al. was in Matlab code. Thus, we decided to reimplement it in Python to preprocess the different datasets in our benchmark. For more information about the procedure, I refer you to Aguirre's original paper [12] and code.

Our MABP procedure follows the following steps:

  1. Identification of each heart cycles in the ABP signal.
  2. Fix the cycle size (limiting and padding)
  3. Exclude cycles out of mean+-(125std).
  4. Compute average cycle.
  5. Extract systolic peak and Dicrotic Notch.

[12] Aguirre, N., Grall-Maës, E., Cymberknop, L. J. & Armentano, R. L. Blood pressure morphology assessment from photoplethysmogram and demographic information using deep learning with attention mechanism. Sensors 21, 2167 (2021). ** https://github.com/AguirreNicolas/PPG2IABP/tree/main