loki2.data.dataclass.wsi

WSI Model for Loki2.

This module defines data structures for working with Whole Slide Images (WSI), including metadata handling, patch processing, and dataset classes.

Module Contents

class loki2.data.dataclass.wsi.WSIMetadata

Metadata container for Whole Slide Image information.

Parameters:

  • name – Name identifier for the WSI.

  • slide_path – Full path to the WSI file.

  • metadata – Dictionary containing additional metadata about the WSI.

name: str
slide_path: str | pathlib.Path
metadata: Dict[str, Any]
class loki2.data.dataclass.wsi.WSI

WSI object

Parameters:

  • name (str) – WSI name

  • patient (str) – Patient name

  • slide_path (Union[str, Path]) – Full path to the WSI file.

  • patched_slide_path (Union[str, Path], optional) – Full path to preprocessed WSI files (patches). Defaults to None.

  • embedding_name (Union[str, Path], optional) – Defaults to None.

  • label (Union[str, int, float, np.ndarray], optional) – Label of the WSI. Defaults to None.

  • logger (logging.logger, optional) – Logger module for logging information. Defaults to None.

name: str
patient: str
slide_path: str | pathlib.Path
patched_slide_path: str | pathlib.Path | None = None
embedding_name: str | pathlib.Path | None = None
label: str | int | float | numpy.ndarray | None = None
logger: logging.Logger | None = None
metadata: Dict[str, Any]
all_patch_metadata: List[Dict[str, Any]]
patches_list: List[str]
patch_transform: Callable | None
load_patch_metadata(patch_name: str) → Dict[str, Any]

Return the metadata of a patch with given name.

This function assumes that metadata path is a subpath of the patches dataset path. The patch name should include the patch suffix (e.g., wsi_1_1.png).

Parameters:

patch_name – Name of patch including suffix.

Returns:

Patch metadata dictionary.

Return type:

Dict[str, Any]

set_patch_transform(transform: Callable) → None

Set the transformation function to process a patch.

Parameters:

transform – Transformation function to apply to patches.

process_patch_image(patch_name: str, transform: Callable | None = None) → Tuple[torch.Tensor, Dict[str, Any]]

Process one patch: Load from disk, apply transformation if needed.

ToTensor is applied automatically. The patch name should include the patch suffix (e.g., wsi_1_1.png).

Parameters:

  • patch_name – Name of patch to load, including patch suffix.

  • transform – Optional patch transformation function.

Returns:

  • torch.Tensor: Patch as torch tensor with shape (3, H, W)

  • Dict[str, Any]: Patch metadata as dictionary

Return type:

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

get_number_patches() → int

Return the number of patches for this WSI

Returns:

number of patches

Return type:

int

get_patches(transform: Callable | None = None) → Tuple[torch.Tensor, List[Dict[str, Any]]]

Get all patches for one image.

Parameters:

transform – Optional patch transformation function.

Returns:

  • torch.Tensor: Patched images with shape (num_patches, 3, height, width)

  • List[Dict[str, Any]]: List of metadata dictionaries for each patch

Return type:

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

load_embedding() → torch.Tensor

Load embedding from subfolder patched_slide_path/embedding/

Raises:

FileNotFoundError – If embedding is not given

Returns:

WSI embedding

Return type:

torch.Tensor

class loki2.data.dataclass.wsi.PatchedWSIInference(wsi_object: WSI, transform: Callable)

Bases: torch.utils.data.Dataset

Inference Dataset for calculating embeddings of a single WSI.

This dataset is wrapped around a WSI object and provides access to patches for inference processing.

Parameters:

  • wsi_object – WSI object containing patch information.

  • transform – Inference transformations to apply to patches.

transform
wsi_object
static collate_batch(batch: List[Tuple[torch.Tensor, Dict[str, Any]]]) → Tuple[torch.Tensor, List[Dict[str, Any]]]

Create a custom batch from a list of patch tuples.

Needed to unpack list of tuples with dictionaries and tensors.

Parameters:

batch – Input batch consisting of a list of tuples (patch, patch-metadata).

Returns:

  • torch.Tensor: Stacked patches with shape (batch_size, 3, patch_size, patch_size)

  • List[Dict[str, Any]]: List of metadata dictionaries

Return type:

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