Designing a Beat Detector

The tutorials on this page demonstrate the processes undertaken to design MSPTDfast, an efficient PPG beat detection algorithm.

Designing MSPTDfast (v.1)

This tutorial demonstrates the processes undertaken to design MSPTDfast (v.1), which was designed using a single dataset. The publication describing this work is available here.

  • Install the PPG-beats toolbox. The usual instructions here are for downloading the latest version of the toolbox, whereas you will need the v.2.0 release to replicate the analysis exactly. This can be downloaded from here.
  • Download the PPG-DaLiA dataset in Matlab format from here.
  • Use the assess_multiple_datasets.m script to run the analysis.
  • During this process a new folder will have been created called proc_data_ppg_dalia_lunch_break. Within this folder you will find files storing the analysis steps, including the file containing the results: the ppg_detect_stats.mat file. Note down the location of this file.
  • Finally, analyse the performance of the different algorithm configuration options by running the msptdfast_cinc_analysis.m script.

This tutorial is demonstrated in the following video:

Designing MSPTDfast (v.2)

This tutorial demonstrates the processes undertaken to design MSPTDfast (v.2), which was designed using multiple development datasets, and then benchmarked against state-of-the-art algorithms on multiple testing datasets. The publication describing this work is available here.

  • Install the PPG-beats toolbox. The usual instructions here are for downloading the latest version of the toolbox, whereas you will need the v.2.2 release to replicate the analysis exactly. This can be downloaded from here.
  • Download the required datasets in Matlab format by following the instructions here. The required datasets are: CapnoBase, BIDMC, MIMIC PERform Training, MIMIC PERform Testing, MIMIC PERform AF, MIMIC PERform Ethnicity, WESAD, and PPG-DaLiA.
  • Use the assess_multiple_datasets.m script to run the analysis. Use one of the following options:
  • To follow the design process: set do_msptdfastv2_design to 1 on lines 633 and 1143 of assess_multiple_datasets.m.
  • To follow the internal validation process: set do_msptdfastv2_internal_validation to 1 on lines 638 and 1149 of assess_multiple_datasets.m.
  • To follow the benchmarking process: set do_msptdfastv2_benchmarking to 1 on lines 643 and 1155 of assess_multiple_datasets.m.
  • During this process a new folder will have been created called proc_data_<dataset name>. Within this folder you will find files storing the analysis steps, including the file containing the results: the ppg_detect_stats.mat file. Note down the location of this file for each dataset.
  • Then, run further analyses using the following additional scripts (adjusting the file paths in the scripts to provide the paths to the files mentioned above):
  • To follow the design analysis: use msptdfastv2_design.m.
  • To follow the internal validation analysis: use msptdfastv2_internal_validation.m.
  • To follow the benchmarking analysis: use msptdfastv2_testing.m. To generate the results for performance at different SNRs, you will need to set: uParams.analysis.sig_qual_tools = {'accel'; 'ppgq'}; on line 152 of assess_beat_detectors.m. You will also require the ppg-quality toolbox.
  • To produce the examples of PPG beat detection challenges (shown in Fig. 7 of the publication), use make_plot_of_ppg_beat_detection_challenges_msptdfastv2.m