loki2.models.cell_segmentation.postprocessing

Postprocessing of Loki2 network output for cell detection.

This module provides CPU-based postprocessing for extracting detected cells from network prediction maps.

Module Contents

class loki2.models.cell_segmentation.postprocessing.DetectionCellPostProcessor(nr_types: int, magnification: Literal[20, 40] = 40, gt: bool = False)

Postprocessor for extracting detected cells from prediction maps.

This class processes network prediction maps to extract detected cells using CPU-based operations.

nr_types

magnification

gt

check_network_output(predictions_: Dict[str, torch.Tensor]) → None

Check if the network output is valid.

Parameters:

predictions –

Network predictions dictionary with required keys: * nuclei_binary_map: Binary nucleus predictions. Shape: (B, H, W, 2) * nuclei_type_map: Type prediction of nuclei.

Shape: (B, H, W, self.nr_types)

• hv_map: Horizontal-Vertical nuclei mapping. Shape: (B, H, W, 2)

Raises:

AssertionError – If predictions dictionary is invalid or missing required keys.

post_process_batch(predictions_: Dict[str, torch.Tensor]) → Tuple[torch.Tensor, List[Dict[str, Any]]]

Post-process a batch of predictions and generate cell dictionaries.

Parameters:

predictions –

Network predictions dictionary with required keys: * nuclei_binary_map: Binary nucleus predictions. Shape: (B, H, W, 2) * nuclei_type_map: Type prediction of nuclei.

Shape: (B, H, W, self.nr_types)

• hv_map: Horizontal-Vertical nuclei mapping. Shape: (B, H, W, 2)

Returns:

• Instance map tensor where each instance has own integer. Shape: (B, H, W)

• List of cell dictionaries, one per image in batch

• List of dictionaries. Each List entry is one image. Each dict contains another dict for each detected nucleus.

  For each nucleus, the following information are returned: “bbox”, “centroid”, “contour”, “type_prob”, “type”

Return type:

Tuple[torch.Tensor, List[Dict[str, Any]]]

post_process_single_image(pred_map: numpy.ndarray) → Tuple[numpy.ndarray, dict[int, dict]]

Process one single image and generate cell dictionary and instance predictions

Parameters:

pred_map (np.ndarray) – Combined output of tp, np and hv branches, in the same order. Shape: (H, W, 4)

Returns:

_description_

Return type:

Tuple[np.ndarray, dict[int, dict]]

loki2.models.cell_segmentation.postprocessing.stack_pred_maps(nuclei_type_map: numpy.ndarray, nuclei_binary_map: numpy.ndarray, hv_map: numpy.ndarray) → numpy.ndarray

Creates the prediction map for HoVer-Net post-processing

Args: nuclei_binary_map:

nuclei_type_map (np.ndarray): Type prediction of nuclei. Shape: (B, H, W, self.num_nuclei_classes,) nuclei_binary_map (np.ndarray): Binary Nucleus Predictions. Shape: (B, H, W, 2) hv_map (np.ndarray): Horizontal-Vertical nuclei mapping. Shape: (B, H, W, 2)

Returns:

A numpy array containing the stacked prediction maps. Shape [B, H, W, 4]

Return type:

np.ndarray