Skip to content

sopa.segmentation.tissue

sopa.segmentation.tissue.hsv_otsu(sdata, image_key=None, level=-1, blur_k=5, open_k=5, close_k=5, drop_threshold=0.01)

Perform WSI tissue segmentation. The resulting ROIs are saved as shapes.

Info

This segmentation method first transforms the image from RBG color space to HSV and then on the basis of the saturation channel, it performs the rest of the steps. As a preprocessing step, a median blurring is applied with an element of size blur_k before the otsu. Then a morphological opening and closing are applied as a prostprocessing step with square elements of size open_k and close_k. Lastly, the connected components with size less than drop_threshold * number_of_pixel_of_the_image are removed, and the rest are converted into polygons.

Parameters:

Name Type Description Default
sdata SpatialData

A SpatialData object representing an H&E image

 required
image_key str | None

Optional key of the H&E image

 None
level int

Level of the multiscale image on which the segmentation will be performed

 -1
blur_k int

The kernel size of the median bluring operation

 5
open_k int

The kernel size of the morphological openning operation

 5
close_k int

The kernel size of the morphological closing operation

 5
drop_threshold int

Segments that cover less area than drop_threshold*100% of the number of pixels of the image will be removed

 0.01

Returns:

Type Description
bool

True if tissue segmentation was successful, else False if no polygon was output.
Source code in sopa/segmentation/tissue.py 
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
def hsv_otsu(
    sdata: SpatialData,
    image_key: str | None = None,
    level: int = -1,
    blur_k: int = 5,
    open_k: int = 5,
    close_k: int = 5,
    drop_threshold: int = 0.01,
) -> bool:
    """Perform WSI tissue segmentation. The resulting ROIs are saved as shapes.

    !!! info
        This segmentation method first transforms the image from RBG color space to HSV and then
        on the basis of the saturation channel, it performs the rest of the steps.
        As a preprocessing step, a median blurring is applied with an element of size `blur_k`
        before the otsu. Then a morphological opening and closing are applied as a prostprocessing
        step with square elements of size `open_k` and `close_k`. Lastly, the connected components
        with size less than `drop_threshold * number_of_pixel_of_the_image` are removed, and the
        rest are converted into polygons.

    Args:
        sdata: A `SpatialData` object representing an H&E image
        image_key: Optional key of the H&E image
        level: Level of the multiscale image on which the segmentation will be performed
        blur_k: The kernel size of the median bluring operation
        open_k: The kernel size of the morphological openning operation
        close_k: The kernel size of the morphological closing operation
        drop_threshold: Segments that cover less area than `drop_threshold`*100% of the number of pixels of the image will be removed

    Returns:
        `True` if tissue segmentation was successful, else `False` if no polygon was output.
    """
    import cv2

    image_key, image = get_item(sdata, "images", image_key)

    if level == 0 or not isinstance(image, DataTree):
        log.warn("Running hsv_otsu on the full image can be slow. We recommend using a DataTree")

    if isinstance(image, DataTree):
        level_keys = list(image.keys())
        image: xr.DataArray = next(iter(image[level_keys[level]].values()))

    thumbnail = np.array(image.transpose("y", "x", "c"))
    thumbnail_hsv = cv2.cvtColor(thumbnail, cv2.COLOR_RGB2HSV)
    thumbnail_hsv_blurred = cv2.medianBlur(thumbnail_hsv[:, :, 1], blur_k)
    _, mask = cv2.threshold(thumbnail_hsv_blurred, 0, 255, cv2.THRESH_OTSU + cv2.THRESH_BINARY)

    mask_open = cv2.morphologyEx(mask, cv2.MORPH_OPEN, np.ones((open_k, open_k), np.uint8))
    mask_open_close = cv2.morphologyEx(
        mask_open, cv2.MORPH_CLOSE, np.ones((close_k, close_k), np.uint8)
    )

    num_labels, labels, stats, _ = cv2.connectedComponentsWithStats(mask_open_close, 4, cv2.CV_32S)

    contours = []
    for i in range(1, num_labels):
        if stats[i, 4] > drop_threshold * np.prod(mask_open_close.shape):
            cc = cv2.findContours(
                np.array(labels == i, dtype="uint8"),
                cv2.RETR_TREE,
                cv2.CHAIN_APPROX_NONE,
            )[0][0]
            c_closed = np.array(list(cc) + [cc[0]])
            contours.extend([c_closed.squeeze()])

    polygons = [Polygon(contour) for contour in contours]

    if not len(polygons):
        log.warn(
            "No polygon has been found after tissue segmentation. Check that there is some tissue in the image, or consider updating the segmentation parameters."
        )
        return False

    _save_tissue_segmentation(sdata, polygons, image_key, image)
    return True