spacr.image_quality

Auditable image-quality screening before segmentation, using raw intensities.

Functions

assess_image(image, settings[, channel_ids])	Measure focus, saturation and nonfinite pixels on unnormalized channels.
ensure_no_retained_measurements(root, rejected)	Refuse exclusions that would leave previously measured fields in reports.
excluded_fields(root)	Read the active saved exclusion policy for downstream field consumers.
filter_batch(batch, filenames, settings)	Remove policy-excluded fields without substituting empty label images.
quality_policy(settings)	Validate a saved per-channel policy without inspecting any image.
screen_fields(root, settings[, paths, channel_ids])	Save a field/channel report and return fields explicitly excluded by policy.

Module Contents

spacr.image_quality.assess_image(image, settings, channel_ids=None)

Measure focus, saturation and nonfinite pixels on unnormalized channels.

Parameters:

• image – YX, YXC or leading-dimensions plus YXC array.

• settings – image_qc_* policy, validated before use.

• channel_ids – optional acquisition-channel labels for stored C planes.

Returns:

one metric/reason record per selected channel. Focus is the best plane’s Laplacian variance in raw intensity units squared, avoiding rejection solely because a z stack includes out-of-focus planes. Saturation uses an explicit acquisition level or integer dtype ceiling; it never uses the brightest observed pixel. Object counts are not read.

Raises:

ValueError – channels, shape or saturation calibration are unavailable.

spacr.image_quality.ensure_no_retained_measurements(root, rejected)

Refuse exclusions that would leave previously measured fields in reports.

Existing results are never deleted. Re-screening an analyzed project needs a fresh project if excluded fields already have measurement rows.

Parameters:

• root – project folder containing measurements/measurements.db.

• rejected – rejected field filenames, also matched without extensions.

Returns:

None if no existing measurement row matches a rejected field.

Raises:

ValueError – a table’s file_name column contains a rejected identity.

spacr.image_quality.excluded_fields(root)

Read the active saved exclusion policy for downstream field consumers.

Parameters:

root – project folder whose qc/image_quality.json is authoritative.

Returns:

excluded NPY basenames, or an empty set when no policy is active.

Raises:

ValueError – a saved report has an unsupported or invalid schema.

spacr.image_quality.filter_batch(batch, filenames, settings)

Remove policy-excluded fields without substituting empty label images.

Parameters:

• batch – normalized image batch with its original field axis.

• filenames – matching basenames in batch order.

• settings – current run settings carrying image_qc_excluded_fields.

Returns:

filtered batch and matching filenames in their original order.

spacr.image_quality.quality_policy(settings)

Validate a saved per-channel policy without inspecting any image.

Parameters:

settings – Mask settings containing image_qc_* options.

Returns:

JSON-serializable policy; empty channel list means every channel.

Raises:

ValueError – unsupported mode, invalid channel or invalid threshold.

spacr.image_quality.screen_fields(root, settings, paths=None, channel_ids=None)

Save a field/channel report and return fields explicitly excluded by policy.

Parameters:

• root – project folder; reports go to qc/image_quality.json and .csv.

• settings – Mask settings; mode off clears a previously active policy.

• paths – optional iterable of raw NPY paths; defaults to root/stack.

• channel_ids – optional acquisition-channel labels for supplied arrays.

Returns:

excluded basenames, empty in report-only or off mode. Inputs are never deleted or rewritten. Report and exact policy precede exclusion.

Raises:

ValueError – enabled screening has no raw fields or invalid calibration.

Nested helpers

_qc_network.QCNet.__init__(self)

Build the two convolutional trunks and the shared head.

spacr/image_quality.py:430

_qc_network.QCNet.forward(self, tiles, thumbnail)

Return class logits for a batch of fields.

spacr/image_quality.py:437

_qc_network.trunk()

Four 3x3 convolution stages shared in shape by both branches.

spacr/image_quality.py:417