scanphyslog2bids

Code to convert Philips physiology files ("SCANPHYSLOG") to the BIDS-format, including the estimation of volume triggers using the logged gradient, volume markers, or by interpolation. It writes out BIDSified physio-files (as *.tsv.gz and associated *.json files). From there on, you can use other software to, for example, estimate RETROICOR/HRV/RVT regressors for nuisance regression. I recommend using the PhysIO toolbox for this (worked really well for me in the past, see image below), but FSL's PNM should also work.

Installation

I recommend installing the package from the master branch using pip:

pip install https://github.com/lukassnoek/scanphyslog2bids/archive/master.zip

Or clone this repository locally and install as follows:

python setup.py install

This package uses Python 3.6 or higher and depends on the following Python packages:

• nibabel
• numpy
• pandas
• matplotlib (optional for plots)
• click (for the CLI)

Usage

This package comes with a Python interface and a CLI. See the code below for a minimal example using the Python interface:

from scanphyslog2bids.core import PhilipsPhysioLog

log_file = 'my_scanphyslog_file.log'
out_dir = '~/my_bids_data'  # where the BIDSified data should be saved
deriv_dir '~/my_bids_data/physio'  # where some QC plots should be saved

# fmri_file is used to extract metadata, such as TR and number of volumes
fmri_file = 'fmri_file_associated_with_scanphyslog.nii.gz' 
fmri_img = nib.load(f)
n_dyns = fmri_img.shape[-1]
tr = np.round(fmri_img.header['pixdim'][4], 3)

# Create PhilipsPhysioLog object with info
phlog = PhilipsPhysioLog(f=log_file, tr=tr, n_dyns=ndyns, sf=496, manually_stopped=False)

# Load in data, do some preprocessing
phlog.load()

# Try to align physio data with scan data, using a particular method
# (either "vol_markers", "gradient_log", or "interpolation")
phlog.align(trigger_method='gradient_log')  # load and find vol triggers

# Write out BIDS files
phlog.to_bids(out_dir)  # writes out .tsv.gz and .json files

# Optional: plot some QC graphs for alignment and actual traces
phlog.plot_alignment(out_dir=deriv_dir)  # plots alignment with gradient
phlog.plot_traces(out_dir=deriv_dir)  # plots cardiac/resp traces

The command line interface can be used as follows:

(base) lukas@uva:~/software/scanphyslog2bids$ scanphyslog2bids --help
Usage: scanphyslog2bids [OPTIONS]

Options:
  --file TEXT           Scanphyslog file to convert (mandatory)
  --sf INTEGER          Sampling rate (optional, default: 496)
  --fmri TEXT           Associated fmri file (optional, assuming ndyns and tr are given)
  --ndyns INTEGER       Number of dynamics/volumes (optional, assuming that fmri is given)
  --tr FLOAT            Repetition time of fmri scan (optional, assuming that fmri is given)
  --manualstop          Was the scan manually stopped? (optional, store True)
  --triggermethod TEXT  Method to detect triggers (optional, default: gradient_log)
  --outdir TEXT         Output directory for BIDS file (optional, default: parent-dir of phys-file)
  --plottraces          Whether to plot the traces (optional, store True)
  --plotalignment       Whether to plot the alignment (optional, store True)
  --derivdir TEXT       Derivatives directory (for plots) (optional, default:
  --help                Show this message and exit.

Output

Apart from the BIDSified SCANPHYSLOG files (i.e., a *.tsv.gz with cardiac/respiratory/volume trigger traces and *.json file), the package allows for creating plots of the physio-scan alignment (using the plot_alignment method or the --plotalignment flag) and the actual respiratory and cardiac traces (using the plot_traces method or the --plottraces flag).

The alignment plot looks similar to the figure below, which visualizes the full gradient trace (if available) with the estimated volume triggers on top (first row), the close-up view of the start and end of the trace (second/third row; because here most issues tend to arise during alignment), and a trace of the number of samples in between (estimated) volume triggers (fourth row). The number of samples in between triggers should generally not deviate more than 2 samples across triggers.

You can also use this plot to detect funky gradients.

The "traces" plot allow you to do some visual QC on the respiratory and cardiac traces:

Advice

This package should work for different "types" of SCANPHYSLOG files, but ideally they contain volume markers (ask your Philips technician to enable this), so you're sure about the actual volume onsets. So, if possible, use the vol_markers "trigger method". However, in my experience, this feature is rarely enabled on Philips scanners.

If volume markers are not available, I recommend using the gradient_log method, which often works quite well, except for when your gradients are really funky (e.g., when you tilt the FOV a lot). For most (2D "ascending", i.e., inferior-superior or vice versa) scans, the "y" gradient direction works best to distill the volume onsets (use which_grad='y' when calling the align method).

If you don't have volume markers and the gradients were not logged (this seems to happen at some scanners), you can use the interpolate trigger method. It works by "interpolating" triggers backwards from the end of the file (specifically from the "end marker"). This is, however, definitely not foolproof, as the "end marker" in SCANPHYSLOG files do not always seem to coincide with the actual offset of the last volume. I have found that this "offset" between the end of the last volume and the end marker may vary from 5 samples (~0.01 seconds) to about 166 samples (~0.332 seconds) depending on your scanner's hardware or software (interface) system. When using interpolation trigger method, you can control the assumed offset with the offset_end_scan parameter in the align method (for which the default is set to 20).

Also, make sure you use the right sampling frequency (the sf parameter). The default is set to 496, which is the sampling frequency for wireless physio recorders. If you use a wired physio recorder, the sampling frequency is (as I've been told) 500 Hz.

Issues

Feel free to submit an issue when encountering issues (of better yet: send a PR if you fixed the bug yourself).