示例#1
0
def get_peak_data():
    """ Function for getting some pregenerated physio data """
    physio = io.load_physio(get_test_data_path('ECG.csv'), fs=1000)
    filt = operations.filter_physio(physio, [5., 15.], 'bandpass')
    peaks = operations.peakfind_physio(filt)

    return peaks
示例#2
0
def test_load_history(tmpdir):
    # get paths of data, new history
    fname = get_test_data_path('ECG.csv')
    temp_history = tmpdir.join('tmp').purebasename

    # make physio object and perform some operations
    phys = io.load_physio(fname, fs=1000.)
    filt = operations.filter_physio(phys, [5., 15.], 'bandpass')

    # save history to file and recreate new object from history
    path = io.save_history(temp_history, filt)
    replayed = io.load_history(path, verbose=True)

    # ensure objects are the same
    assert np.allclose(filt, replayed)
    assert filt.history == replayed.history
    assert filt.fs == replayed.fs
示例#3
0
def test_load_physio():
    # try loading pickle file (from io.save_physio)
    pckl = io.load_physio(get_test_data_path('ECG.phys'), allow_pickle=True)
    assert isinstance(pckl, physio.Physio)
    assert pckl.data.size == 44611
    assert pckl.fs == 1000.
    with pytest.warns(UserWarning):
        pckl = io.load_physio(get_test_data_path('ECG.phys'),
                              fs=500.,
                              allow_pickle=True)
    assert pckl.fs == 500.

    # try loading CSV file
    csv = io.load_physio(get_test_data_path('ECG.csv'))
    assert isinstance(csv, physio.Physio)
    assert np.allclose(csv, pckl)
    assert np.isnan(csv.fs)
    assert csv.history[0][0] == 'load_physio'

    # try loading array
    with pytest.warns(UserWarning):
        arr = io.load_physio(np.loadtxt(get_test_data_path('ECG.csv')))
    assert isinstance(arr, physio.Physio)
    arr = io.load_physio(np.loadtxt(get_test_data_path('ECG.csv')),
                         history=[('np.loadtxt', {
                             'fname': 'ECG.csv'
                         })])
    assert isinstance(arr, physio.Physio)

    # try loading physio object (and resetting dtype)
    out = io.load_physio(arr, dtype='float32')
    assert out.data.dtype == np.dtype('float32')
    assert out.history[0][0] == 'np.loadtxt'
    assert out.history[-1][0] == 'load_physio'
    with pytest.raises(TypeError):
        io.load_physio([1, 2, 3])
示例#4
0
def check_physio(data, ensure_fs=True, copy=False):
    """
    Checks that `data` is in correct format (i.e., `peakdet.Physio`)

    Parameters
    ----------
    data : Physio_like
    ensure_fs : bool, optional
        Raise ValueError if `data` does not have a valid sampling rate
        attribute.
    copy: bool, optional
        Whether to return a copy of the provided data. Default: False

    Returns
    -------
    data : peakdet.Physio
        Loaded physio object

    Raises
    ------
    ValueError
        If `ensure_fs` is set and `data` doesn't have valid sampling rate
    """

    from peakdet.io import load_physio

    if not isinstance(data, physio.Physio):
        data = load_physio(data)
    if ensure_fs and np.isnan(data.fs):
        raise ValueError('Provided data does not have valid sampling rate.')
    if copy is True:
        return new_physio_like(data,
                               data.data,
                               copy_history=True,
                               copy_metadata=True)
    return data
示例#5
0
def test_save_physio(tmpdir):
    pckl = io.load_physio(get_test_data_path('ECG.phys'), allow_pickle=True)
    out = io.save_physio(tmpdir.join('tmp').purebasename, pckl)
    assert os.path.exists(out)
    assert isinstance(io.load_physio(out, allow_pickle=True), physio.Physio)
示例#6
0
def phys2cvr(fname_func,
             fname_co2=None,
             fname_pidx=None,
             fname_roi=None,
             fname_mask=None,
             outdir=None,
             freq=None,
             tr=None,
             trial_len=None,
             n_trials=None,
             highcut=None,
             lowcut=None,
             apply_filter=False,
             run_regression=False,
             lagged_regression=True,
             r2model='full',
             lag_max=None,
             lag_step=None,
             l_degree=0,
             denoise_matrix=[],
             scale_factor=None,
             lag_map=None,
             regr_dir=None,
             run_conv=True,
             quiet=False,
             debug=False):
    """
    Run main workflow of phys2cvr.

    Parameters
    ----------
    fname_func : str or path
        Filename of the functional input (nifti or txt)
    fname_co2 : str or path, optional
        Filename of the CO2 (physiological regressor) timeseries.
        Can be either peakdet's output or a txt file.
        If not declared, phys2cvr will consider the average temporal value
        from the input.
        Default: empty
    fname_pidx : str or path, optional
        Filename of the CO2 (physiological regressor) timeseries' PEAKS.
        Required if CO2 file is a txt AND the convolution step is not skipped.
        If not declared AND the convolution step is not skipped, raises an exception.
        Default: empty
    fname_roi : str or path, optional
        Filename of the roi in a nifti volume.
        If declared, phys2cvr will use these voxels .
        If not, phys2cvr will use a mask, either the declared one or estimated
        from the functional input.
        Ignored if input is a txt file.
        Default: empty
    fname_mask : str or path, optional
        Filename of the mask in a nifti volume.
        If declared, phys2cvr will run only on these voxels.
        If not, phys2cvr will estimate a mask from the functional input.
        Ignored if input is a txt file.
        Default: empty
    outdir : str or path, optional
        Output directory
        Default: the directory where `fname_func` is.
    freq : str, int, or float, optional
        Sample frequency of the CO2 regressor. Required if CO2 input is TXT file.
        If declared with peakdet file, it will overwrite the file frequency.
    tr : str, int, or float, optional
        TR of the timeseries. Required if input is TXT file.
        If declared with nifti file, it will overwrite the file TR.
    trial_len : str or int, optional
        Length of each single trial for tasks that have more than one
        (E.g. BreathHold, CO2 challenges, ...)
        Used to improve cross correlation estimation.
        Default: None
    n_trials : str or int, optional
        Number of trials in the task.
        Default: None
    highcut : str, int, or float, optional
        High frequency limit for filter.
        Required if applying a filter.
        Default: empty
    lowcut : str, int, or float, optional
        Low frequency limit for filter.
        Required if applying a filter.
        Default: empty
    apply_filter : bool, optional
        Apply a Butterworth filter to the functional input.
        Default: False
    run_regression : bool, optional
        Also run the regression step within phys2cvr.
        By default, phys2cvr will only estimate the regressor(s) of interest.
        Default: False
    lagged_regression : bool, optional
        Estimates regressors to run a lagged regression approach.
        If `run_regression` is True, also run the lagged regression.
        Can be turned off.
        Default: True
    r2model : 'full', 'partial' or 'intercept', optional
        Submit the model of the R^2 the regression should return (hence, which
        R^2 model the lag regression should be based on).
        Possible options are 'full', 'partial', 'intercept'.
        See `stats.regression` help to understand them.
        Default: 'full'
    lag_max : int or float, optional
        Limits (both positive and negative) of the temporal area to explore,
        expressed in seconds (i.e. ±9 seconds).
        Default: None
    lag_step : int or float, optional
        Step of the lag to take into account in seconds.
        Default: None
    l_degree : int, optional
        Only used if performing the regression step.
        Highest order of the Legendre polynomial to add to the denoising matrix.
        phys2cvr will add all polynomials up to the specified order
        (e.g. if user specifies 3, orders 0, 1, 2, and 3 will be added).
        Default is 0, which will model only the mean of the timeseries.
    denoise_matrix : list of str(s) or path(s), optional
        Add one or multiple denoising matrices to the regression model.
        Ignored if not performing the regression step.
    scale_factor : str, int, or float, optional
        A scale factor to apply to the CVR map before exporting it.
        Useful when using inputs recorded/stored in Volts that have a meaningful
        unit of measurement otherwise, e.g. (CO2 traces in mmHg).
        V->mmHg: CO2[mmHg] = (Patm-Pvap)[mmHg] * 10*CO2[V]/100[V]
        Default: None
    lag_map : str or path, optional
        Filename of a lag map to get lags from.
        Ignored if not running a lagged-GLM regression.
        Default: None
    regr_dir : str, optional
        Directory containing pre-generated lagged regressors, useful
        to (re-)run a GLM analysis.
        Default: None
    run_conv : bool, optional
        Run the convolution of the physiological trace.
        Can be turned off
        Default: True
    quiet : bool, optional
        Return to screen only warnings and errors.
        Default: False
    debug : bool, optional
        Return to screen more output.
        Default: False

    Raises
    ------
    Exception
        - If functional timeseries is lacking TR and the latter was not specified.
        - If functional nifti file is not at least 4D.
        - If mask was specified but it has different dimensions than the
            functional nifti file.
        - If a file type is not supported yet.
        - If physiological file is a txt file and no peaks were provided.
        - If physiological file is lacking frequency and the latter was not specified.
        - If a lag map was specified but it has different dimensions than the
            functional nifti file.
        - If a lag map was specified, lag_step was not, and the lag map seems
            to have different lag_steps inside.
    """
    # If lagged regression is selected, make sure run_regression is true.
    if lagged_regression:
        run_regression = True
    # Add logger and suff
    if outdir:
        outdir = os.path.abspath(outdir)
    else:
        outdir = os.path.join(os.path.split(fname_func)[0], 'phys2cvr')
    outdir = os.path.abspath(outdir)
    petco2log_path = os.path.join(outdir, 'logs')
    os.makedirs(petco2log_path, exist_ok=True)

    # Create logfile name
    basename = 'phys2cvr_'
    extension = 'tsv'
    isotime = datetime.datetime.now().strftime('%Y-%m-%dT%H%M%S')
    logname = os.path.join(petco2log_path, f'{basename}{isotime}.{extension}')

    # Set logging format
    log_formatter = logging.Formatter(
        '%(asctime)s\t%(name)-12s\t%(levelname)-8s\t%(message)s',
        datefmt='%Y-%m-%dT%H:%M:%S')

    # Set up logging file and open it for writing
    log_handler = logging.FileHandler(logname)
    log_handler.setFormatter(log_formatter)
    sh = logging.StreamHandler()

    if quiet:
        logging.basicConfig(level=logging.WARNING,
                            handlers=[log_handler, sh],
                            format='%(levelname)-10s %(message)s')
    elif debug:
        logging.basicConfig(level=logging.DEBUG,
                            handlers=[log_handler, sh],
                            format='%(levelname)-10s %(message)s')
    else:
        logging.basicConfig(level=logging.INFO,
                            handlers=[log_handler, sh],
                            format='%(levelname)-10s %(message)s')

    version_number = _version.get_versions()['version']
    LGR.info(f'Currently running phys2cvr version {version_number}')
    LGR.info(f'Input file is {fname_func}')

    # Check func type and read it
    func_is_1d = io.check_ext(EXT_1D, fname_func)
    func_is_nifti = io.check_ext(EXT_NIFTI, fname_func)

    # Check that all input values have right type
    tr = io.if_declared_force_type(tr, 'float', 'tr')
    freq = io.if_declared_force_type(freq, 'float', 'freq')
    trial_len = io.if_declared_force_type(trial_len, 'int', 'trial_len')
    n_trials = io.if_declared_force_type(n_trials, 'int', 'n_trials')
    highcut = io.if_declared_force_type(highcut, 'float', 'highcut')
    lowcut = io.if_declared_force_type(lowcut, 'float', 'lowcut')
    lag_max = io.if_declared_force_type(lag_max, 'float', 'lag_max')
    lag_step = io.if_declared_force_type(lag_step, 'float', 'lag_step')
    l_degree = io.if_declared_force_type(l_degree, 'int', 'l_degree')
    if l_degree < 0:
        raise ValueError(
            'The specified order of the Legendre polynomials must be >= 0.')
    scale_factor = io.if_declared_force_type(scale_factor, 'float',
                                             'scale_factor')
    if r2model not in stats.R2MODEL:
        raise ValueError(
            f'R^2 model {r2model} not supported. Supported models '
            f'are {stats.R2MODEL}')

    if func_is_1d:
        if tr:
            func_avg = np.genfromtxt(fname_func)
            LGR.info(f'Loading {fname_func}')
            if apply_filter:
                LGR.info('Applying butterworth filter to {fname_func}')
                func_avg = signal.filter_signal(func_avg, tr, lowcut, highcut)
        else:
            raise NameError('Provided functional signal, but no TR specified! '
                            'Rerun specifying the TR')
    elif func_is_nifti:
        func, dmask, img = io.load_nifti_get_mask(fname_func)
        if len(func.shape) < 4:
            raise ValueError(
                f'Provided functional file {fname_func} is not a 4D file!')
        # Read TR or declare its overwriting
        if tr:
            LGR.warning(f'Forcing TR to be {tr} seconds')
        else:
            tr = img.header['pixdim'][4]

        # Read mask (and mask func) if provided
        if fname_mask:
            _, mask, _ = io.load_nifti_get_mask(fname_mask, is_mask=True)
            if func.shape[:3] != mask.shape:
                raise ValueError(
                    f'{fname_mask} and {fname_func} have different sizes!')
            mask = mask * dmask
            LGR.info(
                f'Masking {os.path.basename(fname_func)} using {os.path.basename(fname_mask)}'
            )
            func = func * mask[..., np.newaxis]
            roiref = os.path.basename(fname_mask)
        else:
            mask = dmask
            LGR.warning(
                f'No mask specified, using any voxel different from 0 in '
                f'{os.path.basename(fname_func)}')
            roiref = os.path.basename(fname_func)

        # Read roi if provided
        if fname_roi:
            _, roi, _ = io.load_nifti_get_mask(fname_roi, is_mask=True)
            if func.shape[:3] != roi.shape:
                raise ValueError(
                    f'{fname_roi} and {fname_func} have different sizes!')
            roi = roi * mask
            roiref = os.path.basename(fname_roi)
        else:
            roi = mask
            LGR.warning(
                f'No ROI specified, using any voxel different from 0 in '
                f'{roiref}')

        if apply_filter:
            LGR.info(f'Obtaining filtered average signal in {roiref}')
            func_filt = signal.filter_signal(func, tr, lowcut, highcut)
            func_avg = func_filt[roi].mean(axis=0)
        else:
            LGR.info(f'Obtaining average signal in {roiref}')
            func_avg = func[roi].mean(axis=0)

    else:
        raise NotImplementedError(
            f'{fname_func} file type is not supported yet, or '
            'the extension was not specified.')

    if fname_co2 == '':
        LGR.info(
            f'Computing "CVR" (approximation) maps using {fname_func} only')
        if func_is_1d:
            LGR.warning(
                'Using an average signal only, solution might be unoptimal.')

            if apply_filter is None:
                LGR.warning('No filter applied to the input average! You know '
                            'what you are doing, right?')

        petco2hrf = func_avg

    else:
        co2_is_phys = io.check_ext('.phys', fname_co2)
        co2_is_1d = io.check_ext(EXT_1D, fname_co2)

        if co2_is_1d:
            if fname_pidx:
                pidx = np.genfromtxt(fname_pidx)
                pidx = pidx.astype(int)
            elif run_conv:
                raise NameError(f'{fname_co2} file is a text file, but no '
                                'file containing its peaks was provided. '
                                ' Please provide peak file!')

            if freq is None:
                raise NameError(f'{fname_co2} file is a text file, but no '
                                'frequency was specified. Please provide peak '
                                ' file!')

            co2 = np.genfromtxt(fname_co2)
        elif co2_is_phys:
            # Read a phys file!
            phys = load_physio(fname_co2, allow_pickle=True)

            co2 = phys.data
            pidx = phys.peaks
            if freq:
                LGR.warning(f'Forcing CO2 frequency to be {freq} Hz')
            else:
                freq = phys.fs
        else:
            raise NotImplementedError(
                f'{fname_co2} file type is not supported yet, or '
                'the extension was not specified.')

        # Set output file & path - calling splitext twice cause .gz
        basename_co2 = os.path.splitext(
            os.path.splitext(os.path.basename(fname_co2))[0])[0]
        outname = os.path.join(outdir, basename_co2)

        # Unless user asks to skip this step, convolve the end tidal signal.
        if run_conv is None or fname_co2 is None:
            petco2hrf = co2
        else:
            petco2hrf = signal.convolve_petco2(co2, pidx, freq, outname)

    # If a regressor directory is not specified, compute the regressors.
    if regr_dir is None:
        regr, regr_shifts = stats.get_regr(func_avg, petco2hrf, tr, freq,
                                           outname, lag_max, trial_len,
                                           n_trials, '.1D', lagged_regression)
    elif run_regression:
        try:
            regr = np.genfromtxt(f'{outname}_petco2hrf.1D')
        except IOError:
            LGR.warning(f'Regressor {outname}_petco2hrf.1D not found. '
                        'Estimating it.')
            regr, regr_shifts = stats.get_regr(func_avg, petco2hrf, tr, freq,
                                               outname, lag_max, trial_len,
                                               n_trials, '.1D')

    # Run internal regression if required and possible!
    if func_is_nifti and run_regression:
        LGR.info('Running regression!')

        # Change dimensions in image header before export
        LGR.info('Prepare output image')
        fname_out_func, _ = io.check_ext('.nii.gz',
                                         os.path.basename(fname_func),
                                         remove=True)
        fname_out_func = os.path.join(outdir, fname_out_func)
        newdim = deepcopy(img.header['dim'])
        newdim[0], newdim[4] = 3, 1
        oimg = deepcopy(img)
        oimg.header['dim'] = newdim

        # Compute signal percentage change of functional data
        m = func.mean(axis=-1)[..., np.newaxis]
        func = (func - m) / m
        func[np.isnan(func)] = 0

        # Start computing the polynomial regressor (at least average)
        LGR.info(f'Compute Legendre polynomials up to order {l_degree}')
        mat_conf = stats.get_legendre(l_degree, regr.size)

        # Read in eventual denoising factors
        if denoise_matrix:
            denoise_matrix = io.if_declared_force_type(denoise_matrix, 'list',
                                                       'denoise_matrix')
            for matrix in denoise_matrix:
                LGR.info(f'Read confounding factor from {matrix}')
                conf = np.genfromtxt(matrix)
                mat_conf = np.hstack([mat_conf, conf])

        LGR.info('Compute simple CVR estimation (bulk shift only)')
        x1D = os.path.join(outdir, 'mat', 'mat_simple.1D')
        beta, tstat, r_square = stats.regression(func, mask, regr, mat_conf,
                                                 r2model, debug, x1D)

        LGR.info('Export bulk shift results')
        if scale_factor is None:
            LGR.warning('Remember: CVR might not be in %BOLD/mmHg!')
        else:
            beta = beta / float(scale_factor)
        # Scale beta by scale factor while exporting (useful to transform V in mmHg)
        LGR.info('Export CVR and T-stat of simple regression')
        io.export_nifti(beta, oimg, f'{fname_out_func}_cvr_simple')
        io.export_nifti(tstat, oimg, f'{fname_out_func}_tstat_simple')

        if debug:
            LGR.debug('Export R^2 volume of simple regression')
            io.export_nifti(r_square, oimg,
                            f'{fname_out_func}_r_square_simple')

        if lagged_regression and regr_shifts is not None and (
            (lag_max and lag_step) or lag_map):
            if lag_max:
                LGR.info(
                    f'Running lagged CVR estimation with max lag = {lag_max}! '
                    '(might take a while...)')
            elif lag_map is not None:
                LGR.info(
                    f'Running lagged CVR estimation with lag map {lag_map}! '
                    '(might take a while...)')

            nrep = int(lag_max * freq * 2)
            if lag_step:
                step = int(lag_step * freq)
            else:
                step = 1

            if regr_dir:
                outprefix = os.path.join(regr_dir, os.path.split(outname)[1])

            # If user specified a lag map, use that one to regress things
            if lag_map:
                lag, _, _ = io.load_nifti_get_mask(lag_map)
                if func.shape[:3] != lag.shape:
                    raise ValueError(
                        f'{lag_map} and {fname_func} have different sizes!')

                # Read lag_step and lag_max from file (or try to)
                lag = lag * mask

                lag_list = np.unique(lag)

                if lag_step is None:
                    lag_step = np.unique(lag_list[1:] - lag_list[:-1])
                    if lag_step.size > 1:
                        raise ValueError(
                            f'phys2cvr found different delta lags in {lag_map}'
                        )
                    else:
                        LGR.warning(
                            f'phys2cvr detected a delta lag of {lag_step} seconds'
                        )
                else:
                    LGR.warning(f'Forcing delta lag to be {lag_step}')

                if lag_max is None:
                    lag_max = np.abs(lag_list).max()
                    LGR.warning(
                        f'phys2cvr detected a max lag of {lag_max} seconds')
                else:
                    LGR.warning(f'Forcing max lag to be {lag_max}')

                lag_idx = (lag + lag_max) * freq / step

                lag_idx_list = np.unique[lag_idx]

                # Prepare empty matrices
                beta = np.empty_like(lag)
                tstat = np.empty_like(lag)

                for i in lag_idx_list:
                    LGR.info(f'Perform L-GLM for lag {lag_list[i]} ({i+1} of '
                             f'{nrep // step})')
                    try:
                        regr = regr_shifts[:, (i * step)]
                    except NameError:
                        regr = np.genfromtxt(f'{outprefix}_{i:04g}')

                    x1D = os.path.join(outdir, 'mat', f'mat_{i:04g}.1D')
                    (beta[lag_idx == i], tstat[lag_idx == i],
                     _) = stats.regression(func[lag_idx == i], [lag_idx == i],
                                           regr, mat_conf, r2model, debug, x1D)

            else:
                # Prepare empty matrices
                r_square_all = np.empty(list(func.shape[:3]) + [nrep // step])
                beta_all = np.empty(list(func.shape[:3]) + [nrep // step])
                tstat_all = np.empty(list(func.shape[:3]) + [nrep // step])

                for n, i in enumerate(range(0, nrep, step)):
                    LGR.info(f'Perform L-GLM number {n+1} of {nrep // step}')
                    try:
                        regr = regr_shifts[:, i]
                        LGR.debug(
                            f'Using shift {i} from matrix in memory: {regr}')
                    except NameError:
                        regr = np.genfromtxt(f'{outprefix}_{i:04g}')
                        LGR.debug(
                            f'Reading shift {i} from file {outprefix}_{i:04g}')

                    x1D = os.path.join(outdir, 'mat', f'mat_{i:04g}.1D')
                    (beta_all[:, :, :, n], tstat_all[:, :, :, n],
                     r_square_all[:, :, :, n]) = stats.regression(
                         func, mask, regr, mat_conf, r2model, debug, x1D)

                if debug:
                    LGR.debug('Export all betas, tstats, and R^2 volumes.')
                    newdim_all = deepcopy(img.header['dim'])
                    newdim_all[0], newdim_all[4] = 4, int(nrep // step)
                    oimg_all = deepcopy(img)
                    oimg_all.header['dim'] = newdim_all
                    io.export_nifti(r_square_all, oimg_all,
                                    f'{fname_out_func}_r_square_all')
                    io.export_nifti(tstat_all, oimg_all,
                                    f'{fname_out_func}_tstat_all')
                    io.export_nifti(beta_all, oimg_all,
                                    f'{fname_out_func}_beta_all')

                # Find the right lag for CVR estimation
                lag_idx = np.argmax(r_square_all, axis=-1)
                lag = (lag_idx * step) / freq - (mask * lag_max)
                # Express lag map relative to median of the roi
                lag_rel = lag - (mask * np.median(lag[roi]))

                # Run through indexes to pick the right value
                lag_idx_list = np.unique(lag_idx)
                beta = np.empty_like(lag)
                tstat = np.empty_like(lag)
                for i in lag_idx_list:
                    beta[lag_idx == i] = beta_all[:, :, :, i][lag_idx == i]
                    tstat[lag_idx == i] = tstat_all[:, :, :, i][lag_idx == i]

            LGR.info('Export fine shift results')
            if scale_factor is None:
                LGR.warning('Remember: CVR might not be in %BOLD/mmHg!')
            else:
                beta = beta / float(scale_factor)

            io.export_nifti(beta, oimg, f'{fname_out_func}_cvr')
            io.export_nifti(tstat, oimg, f'{fname_out_func}_tstat')
            if not lag_map:
                io.export_nifti(lag, oimg, f'{fname_out_func}_lag')
                io.export_nifti(lag_rel, oimg, f'{fname_out_func}_lag_mkrel')

    elif run_regression:
        LGR.warning('The input file is not a nifti volume. At the moment, '
                    'regression is not supported for other formats.')

    LGR.info('phys2cvr finished! Enjoy your outputs!')