Skip to main content

functions

Fluid-volume calculation functions (ported from the v8 algorithm).

These are pure free-function ports of the fluid-volume math that lived on _WorkerSide in bitfount.federated.algorithms.ophthalmology.fluid_volume_calculation_algorithm. The legacy algorithm file is left untouched (additive port); the math here is preserved verbatim (voxel formula, scipy.ndimage.label usage, zeroed-metrics early returns, and rounding).

The low-level convert_nan_to_zero and parse_mask_json primitives are reused from steps.data_utils rather than re-ported. The distance helper is fluid's own 3-axis variant (image centre as fovea proxy) and is deliberately not the 2-axis GA helper.

Module​

Functions​

compute_fluid_metrics_for_scan​

def compute_fluid_metrics_for_scan(    bscan_predictions: tuple[str, ...],    slice_thickness: float,    pixel_spacing_column: float,    pixel_spacing_row: float,    fluid_volume_include_segmentations: list[str],) ‑> FluidVolumeMetrics:

Compute fluid-volume metrics for a single scan.

This encapsulates the per-file computation loop body (steps 6b-6n) from _WorkerSide.run(), parallel to compute_ga_metrics_with_fovea_for_scan.

Raises

  • Exception: Propagates any exception from prediction parsing or metric computation so that the caller can handle/skip.

get_lesion_volumes​

def get_lesion_volumes(    labeled_array: NDArray[Any], num_lesions: int, voxel_volume: float,) ‑> list[float]:

Calculate the volume of each lesion in nL.

Ported verbatim from _WorkerSide._get_lesion_volumes.

Arguments

  • labeled_array: Array of shape (num_bscans, num_rows, num_cols) where each voxel is labelled with the lesion number it belongs to.
  • num_lesions: Number of lesions in the image.
  • voxel_volume: Volume of a single voxel in nL.

Returns List of lesion volumes in nL.

get_max_bscan_index​

def get_max_bscan_index(mask: NDArray[Any], total_volume: float) ‑> int | None:

Return the index of the B-scan with the largest fluid volume.

Ported verbatim from _WorkerSide._get_max_bscan_index.

Arguments

  • mask: Binary mask of fluid across B-scans (shape: num_bscans, num_rows, num_cols).
  • total_volume: Total fluid volume across all B-scans (used to check if any fluid is present).

Returns Index of the B-scan with the largest fluid volume, or None if no fluid.

get_shortest_distance_from_image_centre​

def get_shortest_distance_from_image_centre(    mask: NDArray[Any], voxel_sizes: tuple[float, float, float],) ‑> float:

Distance from the image centre to the nearest lesion voxel (3-axis).

Ported verbatim from _WorkerSide._get_shortest_distance_from_image_centre. Image centre is used as a proxy for the fovea. This is the fluid-specific 3-axis variant and is distinct from the GA 2-axis helper.

Arguments

  • mask: Array of shape (num_bscans, num_rows, num_cols) where each voxel is 1 if part of a lesion, 0 otherwise.
  • voxel_sizes: Per-axis voxel sizes in mm, in the same axis order as mask: (slice_thickness, pixel_spacing_row, pixel_spacing_column).

Returns Distance from the image centre to the nearest lesion voxel in mm, or NaN when the mask has no lesion voxels.

parse_bscan_fluid_predictions_with_cnv​

def parse_bscan_fluid_predictions_with_cnv(    bscan_prediction_strs: tuple[str, ...], fluid_volume_include_segmentations: list[str],) ‑> tuple[dict[str, numpy.ndarray[typing.Any, numpy.dtype[typing.Any]]], list[float]]:

Convert per-B-scan predictions to fluid masks and CNV probabilities.

Ported verbatim from _WorkerSide._parse_bscan_fluid_predictions_with_cnv. Reads the top-level cnv_probability, requires a "mask" key (frames without one are skipped), and reads mask.instances / mask.metadata (via parse_mask_json).

Arguments

  • bscan_prediction_strs: Tuple of predictions for a single scan's B-scans.
  • fluid_volume_include_segmentations: Segmentation labels to build masks for. Falls back to FLUID_VOLUME_INCLUDE_SEGMENTATION_LABELS when empty.

Returns A tuple of:

  • dict mapping each included segmentation label to a stacked binary mask of shape (num_bscans, num_rows, num_cols), or a (0, 0, 0) array when that label had no B-scan masks.
  • list of CNV probabilities (one per parsed B-scan).