histoslice 0.6.0

Read and process histological slide images with python!

HistoSlice

Preprocessing large medical images for machine learning made easy!

Documentation • PyPI

Description

HistoSlice makes is easy to prepare your histological slide images for deep learning models. You can easily cut large slide images into smaller tiles and then preprocess those tiles (remove tiles with shitty tissue, finger marks etc).

[!NOTE] This project was forked from HistoPrep, and further modified for additional features and improvements.

Installation

uv add histoslice
# or
pip install histoslice

Usage

[!NOTE] HistoSlice uses pyvips as the only slide backend. The backend argument is still accepted for compatibility, but it always resolves to pyvips.

If Pillow is built without JPEG support, HistoSlice will automatically save tiles/thumbnails as .png and update filenames accordingly. Developers can check availability via histoslice.functional.has_jpeg_support().

Typical workflow for training deep learning models with histological images is the following:

1. Cut each slide image into smaller tile images.
2. Preprocess smaller tile images by removing tiles with bad tissue, staining artifacts.

histoslice --input './train_images/*.tiff' --output ./tiles --width 512 --overlap 0.5 --max-background 0.5 --backend pyvips --metrics --thumbnail

Or you can use the HistoSlice python API to do the same thing!

from histoslice import SlideReader

# Read slide image.
reader = SlideReader("./slides/slide_with_ink.jpeg")
# Detect tissue.
threshold, tissue_mask = reader.get_tissue_mask(level=-1)
# Extract overlapping tile coordinates with less than 50% background.
tile_coordinates = reader.get_tile_coordinates(
    tissue_mask, width=512, overlap=0.5, max_background=0.5
)
# Save tile images with image metrics for preprocessing.
tile_metadata, failures = reader.save_regions(
    "./train_tiles/",
    tile_coordinates,
    threshold=threshold,
    save_metrics=True,
    save_thumbnail=True
)
if failures:
    print(f"Some tiles failed: {len(failures)}")

Let's take a look at the output and visualise the thumbnails.

train_tiles
└── slide_with_ink
    ├── metadata.parquet       # tile metadata
    ├── failures.json          # per-tile failures (only written if any failures occur)
    ├── properties.json        # tile properties
    ├── thumbnail.jpeg         # thumbnail image (or .png if JPEG support is unavailable)
    ├── thumbnail_tiles.jpeg   # thumbnail with tiles (or .png if JPEG support is unavailable)
    ├── thumbnail_tissue.jpeg  # thumbnail of the tissue mask (or .png if JPEG support is unavailable)
    └── tiles [390 entries exceeds filelimit, not opening dir]

As we can see from the above images, histological slide images often contain areas that we would not like to include into our training data. Might seem like a daunting task but let's try it out!

from histoslice.utils import OutlierDetector

# Let's wrap the tile metadata with a helper class.
detector = OutlierDetector(tile_metadata)
# Cluster tiles based on image metrics.
clusters = detector.cluster_kmeans(num_clusters=4, random_state=666)
# Visualise first cluster.
reader.get_annotated_thumbnail(
    image=reader.read_level(-1), coordinates=detector.coordinates[clusters == 0]
)

Now we can mark tiles in cluster 0 as outliers!