"""Structure containing the parameters defining an ultrasound acquisition.
This module provides the :class:`Parameters` class, a flexible structure
for managing all parameters related to an ultrasound acquisition (merged probe
and scan parameters).
Features
^^^^^^^^
- **Flexible initialization:** The :class:`Parameters` class supports lazy initialization,
allowing you to specify any combination of supported parameters. You can pass only
the parameters you have, and the rest will be computed or set to defaults as needed.
- **Automatic computation:** Many scan properties (such as
grid, number of pixels, wavelength, etc.) are computed automatically from the
provided parameters. This enables you to work with minimal input and still obtain
all necessary scan configuration details.
- **Dependency tracking and lazy evaluation:** Derived properties are computed only
when accessed, and are automatically invalidated and recomputed if their dependencies
change. This ensures efficient memory usage and avoids unnecessary computations.
- **Parameter validation:** All parameters are type-checked and validated against
a predefined schema, reducing errors and improving robustness.
- **Selection of transmits:** The scan supports flexible selection of transmit events,
using the :meth:`set_transmits` method. You can select all, a specific number,
or specific transmit indices. The selection is stored and can be accessed via
the :attr:`selected_transmits` property.
Comparison to ``zea.Config`` and ``zea.Probe``
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
- :class:`zea.config.Config`: A general-purpose parameter dictionary for experiment and pipeline
configuration. It is not specific to ultrasound acquisition and does not compute
derived parameters.
- :class:`zea.probes.Probe`: Contains only probe-specific parameters (e.g., geometry, frequency).
- :class:`zea.Parameters`: Combines all parameters relevant to an ultrasound acquisition,
including probe, acquisition, and scan region. It also provides automatic computation
of derived properties and dependency management.
Example Usage
^^^^^^^^^^^^^
.. doctest::
>>> from zea import Config, File, Probe, Parameters
>>> # The usual entry point: load the merged probe + scan parameters from a file
>>> path = (
... "hf://zeahub/picmus/database/experiments/contrast_speckle/"
... "contrast_speckle_expe_dataset_iq/contrast_speckle_expe_dataset_iq.hdf5"
... )
>>> with File(path) as f:
... parameters = f.load_parameters()
>>> type(parameters).__name__
'Parameters'
>>> # You can also build one from a Probe's parameters ...
>>> probe = Probe.from_name("verasonics_l11_4v")
>>> parameters = Parameters(
... probe_geometry=probe.probe_geometry,
... center_frequency=probe.probe_center_frequency,
... element_width=probe.element_width,
... grid_size_z=256,
... n_tx=11,
... )
>>> # ... from a Config's parameters ...
>>> config = Config.from_path("hf://zeahub/configs/config_picmus_rf.yaml")
>>> parameters = Parameters(n_tx=11, **config.parameters)
>>> # ... or fully manually
>>> parameters = Parameters(
... grid_size_x=128,
... grid_size_z=256,
... xlims=(-0.02, 0.02),
... zlims=(0.0, 0.06),
... ylims=(0.0, 0.0),
... center_frequency=6.25e6,
... sound_speed=1540.0,
... sampling_frequency=25e6,
... n_el=128,
... n_tx=11,
... probe_geometry=probe.probe_geometry,
... )
>>> # Access a derived property (computed lazily)
>>> grid = parameters.grid # shape: (grid_size_z, grid_size_x, 3)
>>> # Select a subset of transmit events
>>> _ = parameters.set_transmits(3) # Use 3 evenly spaced transmits
>>> _ = parameters.set_transmits([0, 2, 4]) # Use specific transmit indices
>>> _ = parameters.set_transmits("all") # Use all transmits
"""
from copy import deepcopy
from typing import Any, ClassVar
import numpy as np
from keras import ops
from zea import log
from zea.beamform.geometry import (
compute_element_normals,
is_diverging,
is_focused,
is_plane_wave,
)
from zea.beamform.pfield import compute_pfield
from zea.beamform.pixelgrid import (
cartesian_pixel_grid,
check_for_aliasing,
polar_pixel_grid,
scanline_aligned_apodization,
scanline_pixel_grid,
)
from zea.data.spec import ProbeSpec, ScanSpec
from zea.display import compute_scan_convert_2d_coordinates
from zea.func.ultrasound import compute_time_to_peak
from zea.internal.parameters import BaseParameters, MissingDependencyError, cache_with_dependencies
from zea.internal.utils import deprecated
from zea.probes import Probe, fit_curved_probe_radius
[docs]
class Parameters(BaseParameters):
"""Contains and computes all parameters relevant to an ultrasound acquisition.
A :class:`Parameters` object holds **all** parameters relevant to an
acquisition — merged probe and scan parameters — and computes derived
quantities (grid, wavelength, pfield, scan-conversion coordinates, ...)
lazily with dependency tracking and caching. Obtain one from a file via
:meth:`zea.data.file.File.load_parameters`.
The set of valid file-backed parameters is derived from
:class:`~zea.data.spec.ScanSpec` and :class:`~zea.data.spec.ProbeSpec`
(single source of truth), extended with recon/beamforming parameters that
are not stored in the file. Arbitrary custom parameters may also be set;
they are stored as-is and are not used to compute any derived parameters
(e.g. the beamforming grid). They are simply passed through — for example
to a pipeline call (see :class:`~zea.internal.parameters.BaseParameters`).
"""
sound_speed: float
"""Speed of sound in the medium [m/s]. Defaults to 1540.0."""
sampling_frequency: float
"""Sampling frequency of the received signal [Hz]."""
center_frequency: float | np.ndarray
"""Transmit center frequency [Hz]. Scalar, or shape (n_tx,) per transmit."""
probe_center_frequency: float
"""Nominal center frequency of the probe [Hz]."""
probe_bandwidth_percent: float
"""Probe bandwidth as a percentage of the center frequency. Defaults to 100.0."""
bandwidth: float
"""Width of the RF band kept by processing [Hz], centered on the
``demodulation_frequency``. Used by :class:`~zea.ops.BandPassFilter` and by
:meth:`check_for_aliasing`."""
n_el: int
"""Number of elements in the transducer array."""
n_ax: int
"""Number of axial samples in the received signal."""
n_ch: int
"""Number of channels (1 for RF, 2 for IQ data)."""
n_frames: int
"""Number of frames in the dataset."""
probe_geometry: np.ndarray
"""Element positions [m], shape (n_el, 3)."""
element_width: float
"""Width of each transducer element [m]."""
element_height: float
"""Height (elevation) of each transducer element [m]."""
selected_transmits: list[int] | None
"""Indices of the currently selected transmit events, or ``None`` when not yet
resolved (e.g. image-only files where ``n_tx`` is unknown).
Assigning to this attribute (or calling :meth:`set_transmits`) accepts any of
the following, which are resolved to concrete indices via :meth:`find_transmits`:
- ``None`` or ``"all"``: use all transmits. If the total transmit count
``n_tx`` is not yet known (image-only files), ``set_transmits(None)`` leaves
this as ``None`` until ``n_tx`` is set.
- ``"center"``: use only the center transmit.
- ``int``: select this many evenly spaced transmits.
- ``list``/``np.ndarray``: use these specific transmit indices.
- ``slice``: use transmits given by the slice (e.g. ``slice(0, 10, 2)``).
"""
pixels_per_wavelength: int
"""Number of grid pixels per wavelength. Defaults to 4."""
grid_type: str
"""Beamforming grid type, ``"cartesian"`` or ``"polar"``. Defaults to
``"cartesian"``. Combine with ``enable_scanline=True`` for line-by-line
(scanline) imaging instead of a shared compounded grid."""
dynamic_range: np.ndarray
"""Dynamic range for image display in dB, shape (2,) as (min_dB, max_dB)."""
resolution: float | None
"""Resolution for scan conversion [mm/pixel]. If None, computed from the image."""
f_number: float
"""F-number of the transducer. Defaults to 1.0."""
fill_value: float
"""Value used for out-of-bounds pixels during scan conversion."""
attenuation_coef: float
"""Attenuation coefficient [dB/(MHz*cm)]. Defaults to 0.0."""
apply_lens_correction: bool
"""Whether to apply lens correction to the transmit delays. Defaults to False."""
focal_region_length: float
"""Full length in meters of the region around the focal plane of focused transmits.
In this region the first-arrival and last-arrival transmit delays are linearly blended,
smoothing the focal-plane transition while preserving the same model outside the region.
This targets the focal-depth artifact described by Rindal et al. (IUS 2018). Defaults to
0.0 (disabled, i.e. conventional first/last-arrival switching everywhere). A typical
starting value is 2e-3 m (corresponding to +/-1e-3 m around the focal plane).
"""
lens_thickness: float
"""Thickness of the lens [m]."""
lens_sound_speed: float
"""Speed of sound in the lens material [m/s]."""
phi_range: tuple | np.ndarray
"""Range of phi angles for 3D imaging [rad], shape (2,)."""
pfield_kwargs: dict
"""Extra keyword arguments for the pressure-field computation.
See :func:`zea.beamform.pfield.compute_pfield`. Defaults to ``{}``.
"""
enable_scanline: bool
"""Whether to beamform in scanline (line-by-line) mode: one pixel column
per transmit (see :func:`~zea.beamform.pixelgrid.scanline_pixel_grid`)
instead of a shared compounded ``grid_type`` grid, paired with
``flat_aligned_apodization`` (fed to :class:`~zea.ops.AlignedApodization`).
Combine with ``grid_type="cartesian"`` for vertical focused columns
(linear-scan geometry) or ``grid_type="polar"`` for steered rays from
each transmit's own origin (sector / phased-array geometry). Defaults to
``False``.
.. note::
This only builds the scanline *grid*. For true classical scanline
beamforming, also construct the beamformer with
``Beamform(enable_aligned_apodization=True)`` so that
``flat_aligned_apodization`` masks each pixel to its owning transmit;
without it every transmit is still compounded onto the
one-column-per-transmit grid.
"""
flat_receive_apodization: np.ndarray
"""Optional custom receive-aperture apodization of shape ``(n_pix, n_el)``.
Per-pixel, per-element (receive-channel) weights, fed to
:class:`~zea.ops.ReceiveApodization` when the beamformer is built with
``Beamform(enable_receive_apodization=True)``. This is the user-supplied
counterpart of the built-in f-number mask
(:func:`~zea.beamform.beamformer.fnumber_mask`) and is applied *in addition*
to it (set ``f_number=0`` to use a fully custom receive apodization alone).
``None`` (default) makes :class:`~zea.ops.ReceiveApodization` a no-op.
Distinct from ``flat_aligned_apodization``, which weights the *transmit*
axis (compounding), not the receive channels.
"""
scan_schema = deepcopy(ScanSpec.SCHEMA)
probe_schema = deepcopy(Probe.SCHEMA)
for key in Probe._NON_PARAMETERS:
probe_schema.pop(key)
# Valid parameters are derived from the scan and probe specs + a few
# beamforming-only parameters.
VALID_PARAMS: ClassVar[dict[str, dict[str, Any]]] = {
**scan_schema,
**probe_schema,
"grid_size_x": {"dtype": np.int32},
"grid_size_y": {"dtype": np.int32},
"grid_size_z": {"dtype": np.int32},
"xlims": {"dtype": np.float32, "shape": (2,)},
"ylims": {"dtype": np.float32, "shape": (2,)},
"zlims": {"dtype": np.float32, "shape": (2,)},
"pixels_per_wavelength": {"dtype": np.int32, "default": 4},
"pfield_kwargs": {"dtype": dict, "default": {}},
"apply_lens_correction": {"dtype": bool, "default": False}, # native dtype on purpose
"enable_scanline": {"dtype": bool, "default": False}, # native dtype on purpose
"flat_aligned_apodization": {"dtype": (type(None), np.ndarray)},
"flat_receive_apodization": {"dtype": (type(None), np.ndarray), "default": None},
"focal_region_length": {"dtype": np.float32, "default": 0.0},
"grid_type": {"dtype": str, "default": "cartesian"},
"polar_limits": {"dtype": np.float32, "shape": (2,)},
"dynamic_range": {"dtype": np.float32, "shape": (2,)},
"selected_transmits": {
"dtype": (type(None), str, int, list, slice, np.ndarray),
"default": None,
},
"n_frames": {"dtype": np.int32},
"n_el": {"dtype": np.int32},
"n_tx": {"dtype": np.int32},
"n_ax": {"dtype": int}, # native dtype on purpose
"n_ch": {"dtype": np.int32},
"bandwidth": {"dtype": np.float32},
"attenuation_coef": {"dtype": np.float32, "default": 0.0},
"f_number": {"dtype": float, "default": 1.0}, # native dtype on purpose
"t_peak": {"dtype": np.float32},
"theta_range": {"dtype": np.float32, "shape": (2,)},
"phi_range": {"dtype": np.float32, "shape": (2,)},
"rho_range": {"dtype": np.float32, "shape": (2,)},
"fill_value": {"dtype": float},
"resolution": {"dtype": (np.float32, type(None)), "default": None},
"distance_to_apex": {"dtype": (np.float32, type(None)), "default": None},
}
# Add some defaults that are not stored in a file
VALID_PARAMS["sound_speed"]["default"] = 1540.0
VALID_PARAMS["probe_bandwidth_percent"]["default"] = 100.0
# Give these a default of None (rather than leaving them unset) so that they can be
# declared as dependencies of computed properties (e.g. t_peak, n_waveforms) without
# tripping MissingDependencyError
VALID_PARAMS["waveforms_one_way"]["default"] = None
VALID_PARAMS["waveforms_two_way"]["default"] = None
@cache_with_dependencies("probe_geometry")
def aperture_size(self):
"""Calculate the aperture size (x,y,z) based on the probe geometry."""
if "probe_geometry" in self._params:
x_coords = self.probe_geometry[:, 0]
y_coords = self.probe_geometry[:, 1]
z_coords = self.probe_geometry[:, 2]
aperture_width = x_coords.max() - x_coords.min()
aperture_height = y_coords.max() - y_coords.min()
aperture_depth = z_coords.max() - z_coords.min()
return np.array([aperture_width, aperture_height, aperture_depth])
return None
@cache_with_dependencies("probe_geometry")
def distance_to_apex(self):
"""The distance from the transducer surface to the apex of the polar pixel grid.
The apex is the point the beams fan out from. For a curved array it is the centre
of curvature, so this is the probe's radius of curvature, fitted from
:attr:`probe_geometry`. A flat array has no centre of curvature and its beams
originate at the array itself, giving 0. Set it explicitly to override the fit.
"""
value = self._params.get("distance_to_apex")
if value is not None:
return value
probe_geometry = self._params.get("probe_geometry")
if probe_geometry is None:
return np.float32(0.0)
try:
return np.float32(fit_curved_probe_radius(probe_geometry))
except ValueError: # not a curved array
return np.float32(0.0)
@cache_with_dependencies(
"xlims",
"ylims",
"zlims",
"grid_size_x",
"grid_size_z",
"grid_size_y",
"grid_type",
"is_3d",
"polar_limits",
"distance_to_apex",
"transmit_origins",
"focus_distances",
"polar_angles",
"azimuth_angles",
"enable_scanline",
)
def grid(self):
"""The beamforming grid of shape (grid_size_z, grid_size_x, [grid_size_y], 3)."""
if self.enable_scanline:
# One pixel column per transmit: scanline imaging is the special case
# of pixel-based DAS where each pixel belongs to exactly one transmit.
# Reuses the same `grid_type` (cartesian/polar) that also picks the
# geometry of the regular (non-scanline) grid below.
polar_angles = self.polar_angles
if polar_angles is None:
log.warning_once(
"No ``polar_angles`` provided, using zeros",
key=(id(self), "polar_angles"),
)
polar_angles = np.zeros(self.n_tx)
return scanline_pixel_grid(
self.transmit_origins,
self.focus_distances,
polar_angles,
self.zlims,
self.grid_size_z,
azimuth_angles=self.azimuth_angles,
grid_type=self.grid_type,
)
elif self.grid_type == "polar":
if self.is_3d:
raise NotImplementedError(
"3D polar grids are not yet supported. Set grid_type='cartesian', "
"or drop ylims/grid_size_y to build a 2D polar grid."
)
return polar_pixel_grid(
self.polar_limits,
self.zlims,
self.grid_size_z,
self.grid_size_x,
self.distance_to_apex,
)
elif self.grid_type == "cartesian":
grid = cartesian_pixel_grid(
self.xlims,
self.zlims,
self.ylims,
grid_size_z=self.grid_size_z,
grid_size_x=self.grid_size_x,
grid_size_y=self.grid_size_y,
)
# Aliasing cannot be assessed without the frequency and probe geometry.
if (
self._params.get("center_frequency") is not None
and self._params.get("probe_geometry") is not None
):
try:
self.check_for_aliasing()
except MissingDependencyError:
pass
return grid
else:
raise ValueError(
f"Unsupported grid type: {self.grid_type}. Supported types are "
"'cartesian' and 'polar'."
)
[docs]
def check_for_aliasing(self, demodulated=True, bandwidth=None) -> list:
"""Warn when the beamforming grid under-samples the data it will carry.
Run by :attr:`grid` for cartesian grids. Whether a pixel pitch
under-samples depends on whether the data reaching the grid is RF or
baseband IQ, which the parameters alone do not tell. By default IQ is
assumed: its axial limit is never stricter than the RF one, so a grid
that fails it aliases either way, and the lateral limit is the same for
both. See :func:`zea.beamform.pixelgrid.aliasing_limits` for the
criteria. Only regular cartesian grids, 2D or 3D, are checked.
Args:
demodulated: Whether the data reaching the grid is baseband IQ.
Pass ``False`` to check against the stricter RF axial limit.
bandwidth: The RF band in Hz. Overrides the ``bandwidth`` parameter;
when neither is set, ``probe_bandwidth_percent`` of the
``center_frequency`` is assumed.
Returns:
list of str: The warnings emitted; empty when the grid is adequate.
"""
if self.grid_type != "cartesian" or self.enable_scanline:
return []
def stored(name):
# Read through the property so the transmit selection applies, but only
# when the value was provided: the fallbacks log a warning that the
# check should not trigger.
return getattr(self, name) if self._params.get(name) is not None else None
focus_distances = stored("focus_distances")
tx_apodizations = stored("tx_apodizations")
if focus_distances is None or tx_apodizations is None:
# Unknown transmit geometry: an empty one makes aliasing_limits assume
# the symmetric transmit cone.
focus_distances, tx_apodizations = np.zeros(0), np.zeros((0, 0))
transmit_origins = stored("transmit_origins")
if transmit_origins is None:
transmit_origins = np.zeros((len(focus_distances), 3))
if bandwidth is None:
bandwidth = self._params.get("bandwidth")
if bandwidth is None:
center_frequency = np.max(np.asarray(self.center_frequency, dtype=float))
bandwidth = self.probe_bandwidth_percent / 100 * center_frequency
return check_for_aliasing(
self.xlims,
self.zlims,
self.grid_size_x,
self.grid_size_z,
self.sound_speed,
self.center_frequency,
self.f_number,
demodulation_frequency=self.demodulation_frequency,
focus_distances=focus_distances,
probe_geometry=self.probe_geometry,
tx_apodizations=tx_apodizations,
polar_angles=self.polar_angles,
transmit_origins=transmit_origins,
demodulated=demodulated,
bandwidth=bandwidth,
azimuth_angles=stored("azimuth_angles"),
ylims=self.ylims if self.is_3d else (0.0, 0.0),
grid_size_y=self.grid_size_y if self.is_3d else 1,
element_width=self._params.get("element_width"),
element_height=self._params.get("element_height"),
)
@cache_with_dependencies("enable_scanline", "n_tx", "grid_size_z", "selected_transmits")
def flat_aligned_apodization(self):
"""Per-pixel, per-transmit compounding apodization weight of shape (n_pix, n_tx).
Can be set explicitly to any weighting, in which case it is stored over
the full transmit axis and read back sliced by ``selected_transmits``.
This requires ``enable_scanline`` to be ``False``: a scanline grid has
one column per *selected* transmit, so its pixel axis is defined by the
transmit selection and a mask stored over the full transmit axis cannot
stay aligned with it.
When left unset, it is the one-hot mask (see
:func:`~zea.beamform.pixelgrid.scanline_aligned_apodization`) that
isolates each pixel's owning transmit if ``enable_scanline`` is ``True``,
and ``None`` otherwise, which makes :class:`~zea.ops.AlignedApodization`
a no-op.
This weights the *transmit* axis (compounding), not the receive channels;
for custom receive-aperture apodization see ``flat_receive_apodization``.
"""
value = self._params.get("flat_aligned_apodization")
if value is not None:
if self.enable_scanline:
raise ValueError(
"``flat_aligned_apodization`` cannot be set explicitly when "
"``enable_scanline`` is True: the scanline grid holds one column per "
"selected transmit, so its pixel axis follows the transmit selection and "
"a mask stored over the full transmit axis cannot stay aligned with it. "
"Set ``enable_scanline=False`` to supply a custom mask."
)
_ = self.n_tx # raises a clear error if no transmit selection is resolved yet
if value.shape[1] != self.n_tx_total:
raise ValueError(
"``flat_aligned_apodization`` is stored over the full transmit axis, so "
f"it must have {self.n_tx_total} columns, got shape {value.shape}."
)
return value[:, self.selected_transmits]
if not self.enable_scanline:
return None
return scanline_aligned_apodization(self.n_tx, self.grid_size_z)
@cache_with_dependencies(
"xlims", "wavelength", "pixels_per_wavelength", "enable_scanline", "n_tx"
)
def grid_size_x(self):
"""Grid width in pixels. For a cartesian grid, this is the lateral (x) pixels in the grid,
set to prevent aliasing if not provided. For a polar grid, this can be thought of as
the number for rays in the polar direction. When ``enable_scanline`` is ``True``, this is
fixed to ``n_tx`` (one column per transmit).
"""
if self.enable_scanline:
return self.n_tx
grid_size_x = self._params.get("grid_size_x")
if grid_size_x is not None:
return grid_size_x
# Pixels sit on both limits, so n pixels span n - 1 pitches.
width = self.xlims[1] - self.xlims[0]
min_grid_size_x = int(np.ceil(width / (self.wavelength / self.pixels_per_wavelength))) + 1
return max(min_grid_size_x, 1)
@cache_with_dependencies(
"ylims",
"wavelength",
"pixels_per_wavelength",
)
def grid_size_y(self):
"""Grid height in pixels. For a cartesian grid, this is the vertical (y) pixels in the grid,
set to prevent aliasing if not provided. For a polar grid, this can be thought of as
the number for rays in the azimuthal direction.
"""
grid_size_y = self._params.get("grid_size_y")
if grid_size_y is not None:
return grid_size_y
height = self.ylims[1] - self.ylims[0]
min_grid_size_y = int(np.ceil(height / (self.wavelength / self.pixels_per_wavelength))) + 1
return max(min_grid_size_y, 1)
@cache_with_dependencies(
"zlims",
"wavelength",
"pixels_per_wavelength",
)
def grid_size_z(self):
"""Grid depth in pixels. This is the number of axial (z) pixels in the grid,
set to prevent aliasing if not provided. When ``enable_scanline`` is ``True``,
this is the number of depth samples per transmit line."""
grid_size_z = self._params.get("grid_size_z")
if grid_size_z is not None:
return grid_size_z
depth = self.zlims[1] - self.zlims[0]
min_grid_size_z = int(np.ceil(depth / (self.wavelength / self.pixels_per_wavelength))) + 1
return max(min_grid_size_z, 1)
@cache_with_dependencies("sound_speed", "center_frequency")
def wavelength(self):
"""Calculate the wavelength based on sound speed and transmit center frequency."""
return self.sound_speed / self.center_frequency
@cache_with_dependencies(
"zlims",
"probe_geometry",
"distance_to_apex",
"focus_distances",
"polar_angles",
"f_number",
"sound_speed",
"sampling_frequency",
"n_ax",
)
def xlims(self):
"""The lateral (x) limits of the beamforming grid in meters.
If not explicitly provided, the limits are derived from the probe geometry, the transmits
and the receive :attr:`f_number`:. If f_number is 0, a 45 degree cone is used instead.
The limits never come out narrower than the probe width.
Unsteered focused or plane wave transmits on a flat array (e.g. walking aperture scans)
image the strip in front of the array, so the limits hug the probe width. Otherwise the
field of view follows the receive cone for the edge elements.
"""
xlims = self._params.get("xlims")
if xlims is not None:
return xlims
aperture_x = self.probe_geometry[:, 0]
left, right = int(np.argmin(aperture_x)), int(np.argmax(aperture_x))
xmin, xmax = float(aperture_x[left]), float(aperture_x[right])
focus_distances = self.focus_distances
polar_angles = self.polar_angles
if polar_angles is None:
polar_angles = np.zeros_like(focus_distances)
# Tolerance: converted data stores a nominally unsteered scan as float noise.
unsteered = np.allclose(polar_angles, 0.0, atol=1e-6) and np.all(
(focus_distances >= 0) | np.isinf(focus_distances)
)
if unsteered and self.distance_to_apex == 0:
return (xmin, xmax)
f_number = float(self.f_number)
half_angle = np.arctan(1 / (2 * f_number)) if f_number > 0 else np.pi / 4
normals = ops.convert_to_numpy(
compute_element_normals(ops.convert_to_tensor(self.probe_geometry))
)
tilt = np.arctan2(normals[:, 0], normals[:, 2])
# Outermost accepted ray of each edge element.
angle = tilt[[left, right]] + [-half_angle, half_angle]
depth = max(self.zlims) - self.probe_geometry[[left, right], 2]
# Stop at the deepest pixel or at the end of the record, whichever is first.
reach = depth * np.tan(np.abs(angle))
max_range = self.sound_speed * float(self.n_ax) / self.sampling_frequency / 2
reach = np.minimum(reach, max_range * np.sin(np.abs(angle)))
reach = np.sign(angle) * reach
return (min(xmin, xmin + float(reach[0])), max(xmax, xmax + float(reach[1])))
@cache_with_dependencies(
"zlims", "grid_type", "azimuth_limits", "probe_geometry", "distance_to_apex"
)
def ylims(self):
"""The y-limits of the beamforming grid [m]. If not explicitly set, it is computed based
on the azimuth limits and probe geometry.
"""
ylims = self._params.get("ylims")
if ylims is not None:
return ylims
# If ylims not set, compute based on azimuth limits and probe geometry
radius = max(self.zlims) + self.distance_to_apex
ylims_azimuth = (
(0.0, 0.0) # avoid numerical imprecision with np.cos(np.pi/2)
if self.azimuth_limits is None or self.azimuth_limits[0] == self.azimuth_limits[1]
else (
radius * np.cos(-np.pi / 2 + self.azimuth_limits[0]),
radius * np.cos(-np.pi / 2 + self.azimuth_limits[1]),
)
)
ylims_plane = (min(self.probe_geometry[:, 1]), max(self.probe_geometry[:, 1]))
ylims = (
min(ylims_azimuth[0], ylims_plane[0]),
max(ylims_azimuth[1], ylims_plane[1]),
)
return ylims
@cache_with_dependencies("sound_speed", "sampling_frequency", "n_ax")
def zlims(self):
"""The z-limits of the beamforming grid [m]."""
zlims = self._params.get("zlims")
if zlims is None:
return [0, self.sound_speed * self.n_ax / self.sampling_frequency / 2]
return zlims
@cache_with_dependencies("grid")
def extent(self):
"""
The extent of the beamforming grid in the format: (xmin, xmax, ymin, ymax, zmin, zmax).
"""
# self.grid holds transducer-frame coordinates for every grid type (polar grids
# place their ray origins at -distance_to_apex), so this needs no correction.
xlims = (self.grid[..., 0].min(), self.grid[..., 0].max())
ylims = (self.grid[..., 1].min(), self.grid[..., 1].max())
zlims = (self.grid[..., 2].min(), self.grid[..., 2].max())
return np.array(
[
xlims[0],
xlims[1],
ylims[0],
ylims[1],
zlims[0],
zlims[1],
]
)
@cache_with_dependencies("extent")
def extent_imshow(self):
"""The extent of the beamforming grid in the format: (xmin, xmax, ymin, ymax, zmin, zmax).
Returns:
np.ndarray: The extent of the beamforming grid in the format (xmin, xmax, zmax, zmin).
This format can be used directly in matplotlib's ``plt.imshow``.
"""
xlims_0, xlims_1, ylims_0, ylims_1, zlims_0, zlims_1 = self.extent
if ylims_0 != ylims_1:
log.warning("Are you sure you want to use 2D imshow extent for a 3D grid?")
return np.array([xlims_0, xlims_1, zlims_1, zlims_0])
@cache_with_dependencies("grid")
def flatgrid(self):
"""The beamforming grid of shape (grid_size_z*grid_size_x*grid_size_y, 3)."""
return self.grid.reshape(-1, 3)
@cache_with_dependencies("grid_size_x", "grid_size_y", "grid_size_z")
def is_3d(self):
"""Whether the scan grid is 3D (True) or 2D (False)."""
return self.grid_size_y > 1 and self.grid_size_x > 1 and self.grid_size_z > 1
@property
def n_tx_total(self):
"""The total number of transmits in the full dataset."""
return self._params["n_tx"]
@cache_with_dependencies("selected_transmits")
def n_tx(self):
"""The number of currently selected transmits."""
if self.selected_transmits is None:
raise ValueError(
'No transmits selected. Call `set_transmits(None/"all"/...)` to '
"resolve a selection (this requires the total transmit count `n_tx` "
"to be set first)."
)
return len(self.selected_transmits)
[docs]
def set_transmits(self, selection):
"""Set the selected transmits based on a selection.
Args:
selection: Specifies which transmits to select:
- None: Use all transmits
- "all": Use all transmits
- "center": Use only the center transmit
- "focused": Use only focused transmits (positive finite
``focus_distances``)
- "diverging": Use only diverging transmits (negative finite
``focus_distances``)
- "plane": Use only plane wave transmits (``np.inf``, the
preferred marker, or ``0``)
- int: Select this many evenly spaced transmits
- list/array: Use these specific transmit indices
- slice: Use transmits specified by the slice (e.g., slice(0, 10, 2))
Returns:
The current instance for method chaining.
"""
if selection is None and self._params.get("n_tx") is None:
# n_tx not yet known (e.g. file with image-only data); store as-is.
idx = None
else:
idx = self.find_transmits(selection)
if len(idx) == 0:
log.warning(f"No transmits found for selection '{selection}'.")
self._params["selected_transmits"] = idx
self._invalidate("selected_transmits")
return self
[docs]
def find_transmits(self, selection) -> list:
"""Find transmit events based on a selection.
This method provides flexible ways to select transmit events:
Args:
selection: Specifies which transmits to select:
- None: Use all transmits
- "all": Use all transmits
- "center": Use only the center transmit
- "focused": Use only focused transmits (positive finite
``focus_distances``)
- "diverging": Use only diverging transmits (negative finite
``focus_distances``)
- "plane": Use only plane wave transmits (``np.inf``, the
preferred marker, or ``0``)
- int: Select this many evenly spaced transmits
- list/array: Use these specific transmit indices
- slice: Use transmits specified by the slice (e.g., slice(0, 10, 2))
Returns:
The selected transmit indices.
Raises:
ValueError: If the selection is invalid or incompatible with the scan.
"""
n_tx_total = self._params.get("n_tx")
if n_tx_total is None:
raise ValueError("n_tx must be set.")
# Handle array-like - convert to list of indices
if isinstance(selection, np.ndarray):
if len(selection.shape) == 0:
# Handle scalar numpy array
selection = int(selection)
elif len(selection.shape) == 1:
selection = selection.tolist()
else:
raise ValueError(f"Invalid array shape: {selection.shape}")
# Handle None and "all" - use all transmits
if selection is None or selection == "all":
return list(range(n_tx_total))
# Handle "center" - use center transmit
if selection == "center":
return [n_tx_total // 2]
if selection == "focused":
value = self._params.get("focus_distances")
if value is None:
raise ValueError("No focus distances provided, cannot select focused transmits")
return np.where(is_focused(value))[0].tolist()
if selection == "diverging":
value = self._params.get("focus_distances")
if value is None:
raise ValueError("No focus distances provided, cannot select diverging transmits")
return np.where(is_diverging(value))[0].tolist()
if selection == "plane":
value = self._params.get("focus_distances")
if value is None:
raise ValueError("No focus distances provided, cannot select plane wave transmits")
return np.where(is_plane_wave(value))[0].tolist()
# Handle integer - select evenly spaced transmits
if isinstance(selection, (int, np.integer)):
selection = int(selection) # Convert numpy integer to Python int
if selection <= 0:
raise ValueError(f"Number of transmits must be positive, got {selection}")
if selection > n_tx_total:
raise ValueError(
f"Requested {selection} transmits exceeds available transmits ({n_tx_total})"
)
if selection == 1:
return [n_tx_total // 2]
else:
# Compute evenly spaced indices
tx_indices = np.linspace(0, n_tx_total - 1, selection)
return list(np.rint(tx_indices).astype(int))
# Handle slice - convert to list of indices
if isinstance(selection, slice):
selection = list(range(n_tx_total))[selection]
# Handle list of indices
if isinstance(selection, list):
# Validate indices
if not all(isinstance(i, (int, np.integer)) for i in selection):
raise ValueError("All transmit indices must be integers")
if any(i < 0 or i >= n_tx_total for i in selection):
raise ValueError(f"Transmit indices must be between 0 and {n_tx_total - 1}")
# Convert numpy integers to Python ints
return [int(i) for i in selection]
raise ValueError(f"Unsupported selection type: {type(selection)}")
@cache_with_dependencies("center_frequency")
def demodulation_frequency(self):
"""The demodulation frequency in Hz."""
if self._params.get("demodulation_frequency") is not None:
return self._params["demodulation_frequency"]
return self.center_frequency
@cache_with_dependencies("center_frequency", "probe_bandwidth_percent")
def bandwidth(self):
"""The signal bandwidth in Hz, used by filter operations such as
:class:`~zea.ops.LowPassFilterIQ` and :class:`~zea.ops.BandPassFilter`.
If not set explicitly, it is derived from the probe metadata as
``center_frequency * probe_bandwidth_percent / 100``. When the file does not store
``probe_bandwidth_percent``, its default applies.
The filter operations design one set of taps, so they need a scalar. A per-transmit
``center_frequency`` that is the same for every transmit gives a scalar bandwidth; when
it varies across transmits, no bandwidth is derived (``None``), and ``bandwidth`` must be
set explicitly for these filters.
"""
if self._params.get("bandwidth") is not None:
return self._params["bandwidth"]
bandwidth = self.center_frequency * self.probe_bandwidth_percent / 100
if np.ndim(bandwidth) > 0:
per_transmit = np.asarray(bandwidth)
if per_transmit.size == 0 or np.any(per_transmit != per_transmit.flat[0]):
return None
return per_transmit.flat[0]
return bandwidth
@cache_with_dependencies("selected_transmits")
def polar_angles(self):
"""Polar angles for each transmit event in radians of shape (n_tx,).
These angles are often used in 2D imaging."""
value = self._params.get("polar_angles")
if value is None:
return None
return value[self.selected_transmits]
@cache_with_dependencies("polar_angles")
def polar_limits(self):
"""The limits of the polar angles, used for polar grids."""
value = self._params.get("polar_limits")
if value is None and self.polar_angles is not None:
value = self.polar_angles.min(), self.polar_angles.max()
diff = value[1] - value[0]
# add 15% margin to the limits
value = (value[0] - 0.15 * diff, value[1] + 0.15 * diff)
return value
@cache_with_dependencies("selected_transmits", "n_tx")
def azimuth_angles(self):
"""Azimuth angles for each transmit event in radians
of shape (n_tx,). These angles are often used in 3D imaging."""
value = self._params.get("azimuth_angles")
if value is None:
log.warning_once(
"No ``azimuth_angles`` provided, using zeros",
key=(id(self), "azimuth_angles"),
)
return np.zeros(self.n_tx)
return value[self.selected_transmits]
@cache_with_dependencies("selected_transmits", "n_tx")
def azimuth_limits(self):
"""The limits of the azimuth angles.
Returns ``None`` if azimuth angles were
not provided (e.g. standard 2D imaging).
The presence check uses the raw parameter to avoid triggering the
:attr:`azimuth_angles` "using zeros" fallback warning for 2D data.
"""
value = self._params.get("azimuth_limits")
if value is None and self._params.get("azimuth_angles") is not None:
azimuth_angles = self.azimuth_angles
value = azimuth_angles.min(), azimuth_angles.max()
diff = value[1] - value[0]
# add 15% margin to the limits
value = (value[0] - 0.15 * diff, value[1] + 0.15 * diff)
return value
@cache_with_dependencies("selected_transmits", "n_el", "n_tx")
def t0_delays(self):
"""Transmit delays in seconds of
shape (n_tx, n_el), shifted such that the smallest delay is 0."""
value = self._params.get("t0_delays")
if value is None:
log.warning_once(
"No ``t0_delays`` provided, using zeros",
key=(id(self), "t0_delays"),
)
return np.zeros((self.n_tx, self.n_el))
return value[self.selected_transmits]
@cache_with_dependencies("selected_transmits", "n_el", "n_tx")
def tx_apodizations(self):
"""Transmit apodizations of shape (n_tx, n_el)."""
value = self._params.get("tx_apodizations")
if value is None:
log.warning_once(
"No ``tx_apodizations`` provided, using ones",
key=(id(self), "tx_apodizations"),
)
return np.ones((self.n_tx, self.n_el))
return value[self.selected_transmits]
@cache_with_dependencies("selected_transmits")
def rx_aperture_indices(self):
"""Probe element index of each receive channel of shape (n_tx, n_rx).
``None`` for the usual case where every element receives on every transmit,
in which case the mapping is the identity and :attr:`n_rx` equals
:attr:`n_el`. Only acquisitions with a receive sub-aperture (a sliding
receive window, or a multiplexed front-end with fewer channels than
elements) carry it, since the mapping is then not recoverable from the
channel data alone. Sliced by :attr:`selected_transmits`, so it stays
aligned with :attr:`t0_delays` and :attr:`tx_apodizations`.
The transmit counterpart needs no such field: the transmit aperture is
recorded per element in :attr:`tx_apodizations`, whose zeros mark the
elements that did not fire.
"""
value = self._params.get("rx_aperture_indices")
if value is None:
return None
return value[self.selected_transmits]
@cache_with_dependencies("rx_aperture_indices", "n_el")
def n_rx(self):
"""Number of receive channels in the channel data.
Equal to :attr:`n_el` unless the acquisition used a receive sub-aperture,
which :attr:`rx_aperture_indices` then maps onto the probe's elements.
"""
indices = self.rx_aperture_indices
if indices is None:
return self.n_el
return int(indices.shape[1])
@cache_with_dependencies("selected_transmits", "n_tx")
def focus_distances(self):
"""Focus distances in meters for each event of shape (n_tx,).
The sign and magnitude encode the transmit type: a positive finite value
is a focused transmit (focal point that distance in front of the array),
a negative finite value is a diverging transmit (virtual source that
distance behind the array), and ``np.inf`` is a plane wave. ``np.inf`` is
the preferred, canonical plane-wave value in zea; ``0.0`` is also accepted
(see :meth:`find_transmits`). When no focus distances are provided, zeros
are used (i.e. treated as plane waves).
"""
value = self._params.get("focus_distances")
if value is None:
log.warning_once(
"No ``focus_distances`` provided, using zeros",
key=(id(self), "focus_distances"),
)
return np.zeros(self.n_tx)
return value[self.selected_transmits]
@cache_with_dependencies("selected_transmits", "n_tx")
def transmit_origins(self):
"""Transmit origins of shape (n_tx, 3)."""
value = self._params.get("transmit_origins")
if value is None:
log.warning_once(
"No ``transmit_origins`` provided, using zeros",
key=(id(self), "transmit_origins"),
)
return np.zeros((self.n_tx, 3))
return value[self.selected_transmits]
@cache_with_dependencies("selected_transmits", "n_tx")
def initial_times(self):
"""Initial times in seconds for each event of shape (n_tx,)."""
value = self._params.get("initial_times")
if value is None:
log.warning_once(
"No ``initial_times`` provided, using zeros",
key=(id(self), "initial_times"),
)
return np.zeros(self.n_tx)
return value[self.selected_transmits]
@cache_with_dependencies("selected_transmits")
def waveforms_one_way(self):
"""One-way transmit waveforms of shape (n_tx, n_samples_one_way), or None."""
value = self._params.get("waveforms_one_way")
if value is None:
return None
return value[self.selected_transmits]
@cache_with_dependencies("selected_transmits")
def waveforms_two_way(self):
"""Two-way transmit waveforms of shape (n_tx, n_samples_two_way), or None."""
value = self._params.get("waveforms_two_way")
if value is None:
return None
return value[self.selected_transmits]
@cache_with_dependencies("waveforms_one_way", "waveforms_two_way")
def n_waveforms(self):
"""The number of transmit waveforms, one per selected transmit."""
if self.waveforms_one_way is not None:
return self.waveforms_one_way.shape[0]
if self.waveforms_two_way is not None:
return self.waveforms_two_way.shape[0]
return 1
@cache_with_dependencies("center_frequency", "selected_transmits", "waveforms_two_way")
def t_peak(self):
"""The time of the peak of the pulse in seconds of shape (n_tx,).
If not set explicitly and ``waveforms_two_way`` (the two-way,
pulse-echo transmit waveform) is available, this is estimated from it
via :func:`~zea.func.ultrasound.compute_time_to_peak`. Otherwise
it defaults to ``1 / center_frequency``.
"""
t_peak = self._params.get("t_peak")
if t_peak is None:
waveforms = self._params.get("waveforms_two_way")
if waveforms is not None:
t_peak = ops.convert_to_numpy(
compute_time_to_peak(waveforms, self.center_frequency)
)
else:
t_peak = np.full(self.n_tx_total, 1 / self.center_frequency)
return t_peak[self.selected_transmits]
@cache_with_dependencies("selected_transmits")
def time_to_next_transmit(self):
"""The time between subsequent transmit events of shape (n_frames, n_tx)."""
value = self._params.get("time_to_next_transmit")
if value is None:
return None
return value[:, self.selected_transmits]
@cache_with_dependencies("n_ax")
def tgc_gain_curve(self):
"""Time gain compensation (TGC) curve of shape (n_ax,)."""
value = self._params.get("tgc_gain_curve")
if value is None:
log.warning_once(
"No ``tgc_gain_curve`` provided, using ones",
key=(id(self), "tgc_gain_curve"),
)
return np.ones(self.n_ax)
return value[: self.n_ax]
@cache_with_dependencies(
"sound_speed",
"center_frequency",
"probe_bandwidth_percent",
"probe_geometry",
"tx_apodizations",
"grid",
"t0_delays",
"pfield_kwargs",
"attenuation_coef",
optional=("element_width",),
)
def pfield(self) -> np.ndarray:
"""Compute or return the pressure field (pfield) for weighting
of shape (n_tx, grid_size_z, grid_size_x).
Uses :attr:`attenuation_coef`, unless overridden in :attr:`pfield_kwargs`."""
if self.is_3d:
raise NotImplementedError("Pfield computation is not yet implemented for 3D scans.")
pfield = compute_pfield(
sound_speed=self.sound_speed,
center_frequency=self.center_frequency,
probe_geometry=self.probe_geometry,
tx_apodizations=self.tx_apodizations,
grid=self.grid,
t0_delays=self.t0_delays,
probe_bandwidth_percent=self.probe_bandwidth_percent,
element_width=self._params.get("element_width"),
**{"attenuation_coef": self.attenuation_coef, **self.pfield_kwargs},
)
return ops.convert_to_numpy(pfield)
@cache_with_dependencies("pfield")
def flat_pfield(self):
"""Flattened pfield for weighting of shape (n_pix, n_tx)."""
return self.pfield.reshape(self.n_tx, -1).swapaxes(0, 1)
@cache_with_dependencies("zlims", "distance_to_apex")
def rho_range(self):
"""A tuple specifying the range of rho values (min_rho, max_rho). Defined in meters.
Used for scan conversion. Rho is measured from the apex of the polar grid, so
:attr:`zlims` (depths below the transducer) are shifted by
:attr:`distance_to_apex`, matching the ``rlims`` of
:func:`~zea.beamform.pixelgrid.polar_pixel_grid`."""
value = self._params.get("rho_range")
if value is None:
return (
self.zlims[0] + self.distance_to_apex,
self.zlims[1] + self.distance_to_apex,
)
return value
@cache_with_dependencies("polar_limits")
def theta_range(self):
"""A tuple specifying the range of theta values (min_theta, max_theta).
Defined in radians. Used for scan conversion."""
value = self._params.get("theta_range")
if value is None and self.polar_limits is not None:
return self.polar_limits
return value
@cache_with_dependencies(
"rho_range",
"theta_range",
"resolution",
"grid_size_z",
"grid_size_x",
)
def coordinates_2d(self):
"""The coordinates for scan conversion.
These index the polar image, so they live in the apex frame and need no
``distance_to_apex`` correction; :attr:`rho_range` already carries it.
"""
coords, _ = compute_scan_convert_2d_coordinates(
(self.grid_size_z, self.grid_size_x),
self.rho_range,
self.theta_range,
self.resolution,
)
return coords
@cache_with_dependencies("is_3d", "coordinates_2d")
def coordinates(self):
"""Get the coordinates for scan conversion."""
if self.is_3d:
raise NotImplementedError(
"Scan conversion of 3D grids is not supported. Use "
"zea.display.scan_convert_3d directly on the volume."
)
return self.coordinates_2d
@cache_with_dependencies("time_to_next_transmit")
def pulse_repetition_frequency(self):
"""The pulse repetition frequency (PRF) [Hz]. Assumes a constant PRF."""
if self.time_to_next_transmit is None:
raise ValueError(
"Time to next transmit must be set to compute pulse repetition frequency"
)
pulse_repetition_interval = np.mean(self.time_to_next_transmit)
return 1 / pulse_repetition_interval
@cache_with_dependencies("time_to_next_transmit")
def frames_per_second(self):
"""The number of frames per second [Hz]. Assumes a constant frame rate.
Frames per second computed based on time between transmits within a frame.
Ignores time between frames (e.g. due to processing).
Uses the time it took to do all transmits (per frame). So if you only use some portion
of the transmits, the fps will still be calculated based on all.
"""
time_to_next_transmit = self._params.get("time_to_next_transmit")
if time_to_next_transmit is None:
raise ValueError("Time to next transmit must be set to compute frames per second")
# Check if fps is constant
uniq = np.unique(time_to_next_transmit, axis=0) # frame axis
if uniq.shape[0] != 1:
log.warning("Time to next transmit is not constant")
# Compute fps
time = np.mean(np.sum(time_to_next_transmit, axis=1))
fps = 1 / time
return fps
[docs]
def to_scan_dict(self) -> dict:
"""Return scan parameters as a plain dict.
Suitable for passing directly to :meth:`~zea.File.create` as the
``scan`` argument, or to :func:`~zea.data.file_operations.save_file`
alongside :meth:`to_probe_dict`.
Only fields defined in :class:`~zea.data.spec.ScanSpec` that are
currently stored on this object are included (``None`` values are
omitted). Values are read through property access so that any active
:attr:`selected_transmits` filtering is applied (e.g. after calling
:meth:`set_transmits`).
Returns:
dict: Scan parameter dict keyed by :class:`~zea.data.spec.ScanSpec`
field names.
"""
result = {}
for field in ScanSpec.fields():
# Only include fields that were explicitly stored as a non-None value.
# Skipping None here is critical: fields like ``tgc_gain_curve`` that
# are absent in the source file are stored as None in ``_params``.
# Calling ``getattr`` on them would trigger computed defaults (e.g.
# ones), which would be written to the new file and break round-trips.
if field not in self._params or self._params[field] is None:
continue
try:
value = getattr(self, field) # applies selected_transmits filtering
except MissingDependencyError:
continue
if value is not None:
result[field] = value
return result
[docs]
def to_probe_dict(self) -> dict:
"""Return file-backed probe parameters as a plain dict.
Suitable for passing directly to :meth:`~zea.File.create` as the
``probe`` argument, or to :func:`~zea.data.file_operations.save_file`
alongside :meth:`to_scan_dict`.
Only fields defined in :class:`~zea.data.spec.ProbeSpec` that are
currently stored on this object are included (``None`` values are
omitted).
Returns:
dict: Probe parameter dict keyed by :class:`~zea.data.spec.ProbeSpec`
field names.
"""
return {
field: self._params[field]
for field in ProbeSpec.fields()
if field in self._params and self._params[field] is not None
}
def __setattr__(self, name: str, value):
if name == "selected_transmits":
# If setting selected_transmits, call set_transmits to handle logic
self.set_transmits(value)
else:
return super().__setattr__(name, value)
[docs]
class Scan(Parameters):
"""Deprecated alias for :class:`Parameters`.
``Scan`` was renamed to :class:`zea.Parameters` (which now holds the merged
probe and scan parameters). This subclass is kept temporarily to ease the
transition: instantiating it emits a :class:`DeprecationWarning` pointing to
:class:`zea.Parameters`. It will be removed in a future release.
"""
@deprecated(replacement="zea.Parameters")
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)