data/tests/preprocessing_tests/test_transforms.py

"""
Copyright 2021, Dana-Farber Cancer Institute and Weill Cornell Medicine
License: GNU GPL 2.0
"""

import numpy as np
import pytest

from pathml.core import Tile
from pathml.preprocessing import (
    AdaptiveHistogramEqualization,
    BinaryThreshold,
    BoxBlur,
    CollapseRunsCODEX,
    CollapseRunsVectra,
    ForegroundDetection,
    GaussianBlur,
    HistogramEqualization,
    LabelArtifactTileHE,
    LabelWhiteSpaceHE,
    MedianBlur,
    MorphClose,
    MorphOpen,
    NucleusDetectionHE,
    QuantifyMIF,
    RescaleIntensity,
    StainNormalizationHE,
    SuperpixelInterpolation,
    TissueDetectionHE,
)
from pathml.utils import RGB_to_GREY


@pytest.mark.parametrize("ksize", [3, 7, 21])
@pytest.mark.parametrize("transform", [MedianBlur, BoxBlur])
def test_median_box_blur(tileHE, ksize, transform):
    t = transform(kernel_size=ksize)
    orig_im = tileHE.image
    t.apply(tileHE)
    assert np.array_equal(tileHE.image, t.F(orig_im))


@pytest.mark.parametrize("ksize", [3, 7, 21])
@pytest.mark.parametrize("sigma", [0.1, 3, 0.999])
def test_gaussian_blur(tileHE, ksize, sigma):
    t = GaussianBlur(kernel_size=ksize, sigma=sigma)
    orig_im = tileHE.image
    t.apply(tileHE)
    assert np.array_equal(tileHE.image, t.F(orig_im))


@pytest.mark.parametrize("in_range", ["image", (0, 255), "dtype"])
@pytest.mark.parametrize("out_range", ["image", (0, 255), "dtype"])
def test_rescale_intensity(tileVectra, in_range, out_range):
    t = RescaleIntensity(in_range=in_range, out_range=out_range)
    orig_im = tileVectra.image
    t.apply(tileVectra)
    assert np.array_equal(tileVectra.image, t.F(orig_im))


@pytest.mark.parametrize("nbins", [120, 255, 500])
def test_histogram_equalization(tileVectra, nbins):
    t = HistogramEqualization(nbins=nbins)
    orig_im = tileVectra.image
    t.apply(tileVectra)
    assert np.array_equal(tileVectra.image, t.F(orig_im))


@pytest.mark.parametrize("clip_limit", [0.05, 0.1, 0.3])
@pytest.mark.parametrize("nbins", [120, 255, 500])
def test_adaptive_histogram_equalization(tileVectra, clip_limit, nbins):
    t = AdaptiveHistogramEqualization(clip_limit=clip_limit, nbins=nbins)
    vectra_collapse = CollapseRunsVectra()
    vectra_collapse.apply(tileVectra)
    orig_im = tileVectra.image
    t.apply(tileVectra)
    assert np.array_equal(tileVectra.image, t.F(orig_im))


@pytest.mark.parametrize("thresh", [0, 0.5, 200])
@pytest.mark.parametrize("otsu", [True, False])
def test_binary_thresholding(tileHE, thresh, otsu):
    t = BinaryThreshold(use_otsu=otsu, threshold=thresh, mask_name="testing")
    t.apply(tileHE)
    assert np.array_equal(tileHE.masks["testing"], t.F(RGB_to_GREY(tileHE.image)))


@pytest.mark.parametrize("n_iter", [1, 3])
@pytest.mark.parametrize("ksize", [3, 21])
@pytest.mark.parametrize("transform", [MorphOpen, MorphClose])
def test_open_close(tileHE, transform, ksize, n_iter):
    t = transform(kernel_size=ksize, n_iterations=n_iter, mask_name="testmask")
    orig_mask = np.copy(tileHE.masks["testmask"])
    m = t.F(orig_mask)
    t.apply(tileHE)
    assert np.array_equal(tileHE.masks["testmask"], m)


@pytest.mark.parametrize("min_reg_size", [0, 10])
@pytest.mark.parametrize("max_hole_size", [0, 10])
@pytest.mark.parametrize("outer_contours_only", [True, False])
def test_foreground_detection(tileHE, min_reg_size, max_hole_size, outer_contours_only):
    t = ForegroundDetection(
        min_region_size=min_reg_size,
        max_hole_size=max_hole_size,
        outer_contours_only=outer_contours_only,
        mask_name="testmask",
    )
    orig_mask = tileHE.masks["testmask"]
    t.apply(tileHE)
    assert np.array_equal(tileHE.masks["testmask"], t.F(orig_mask))


@pytest.mark.parametrize("n_iter", [1, 30])
@pytest.mark.parametrize("region_size", [10, 20])
def test_superpix_interp(tileHE, region_size, n_iter):
    t = SuperpixelInterpolation(region_size=region_size, n_iter=n_iter)
    orig_im = tileHE.image
    t.apply(tileHE)
    assert np.array_equal(tileHE.image, t.F(orig_im))


@pytest.mark.parametrize("target", ["normalize", "hematoxylin", "eosin"])
@pytest.mark.parametrize("method", ["vahadane", "macenko"])
def test_stain_normalization_he(tileHE, method, target):
    t = StainNormalizationHE(target=target, stain_estimation_method=method)
    orig_im = tileHE.image
    t.fit_to_reference(orig_im)
    assert t.stain_matrix_target_od.shape == (2, 3)


@pytest.mark.parametrize("target", ["normalize", "hematoxylin", "eosin"])
@pytest.mark.parametrize("method", ["vahadane", "macenko"])
def test_stain_normalization_he_fit(tileHE, method, target):
    t = StainNormalizationHE(target=target, stain_estimation_method=method)
    orig_im = tileHE.image
    t.apply(tileHE)
    if method == "vahadane":
        # theres an element of randomness in vahadane implementation, haven't been able to figure
        # out how to set a seed
        assert tileHE.image.shape == t.F(orig_im).shape
    else:
        assert np.allclose(tileHE.image, t.F(orig_im))


def test_nuc_detectionHE(tileHE):
    t = NucleusDetectionHE(mask_name="testing", stain_estimation_method="macenko")
    orig_im = tileHE.image
    t.apply(tileHE)
    assert np.array_equal(tileHE.masks["testing"], t.F(orig_im))


@pytest.mark.parametrize("use_saturation", [True, False])
@pytest.mark.parametrize("threshold", [None, 100])
def test_tissue_detectionHE(tileHE, threshold, use_saturation):
    t = TissueDetectionHE(
        mask_name="testing", threshold=threshold, use_saturation=use_saturation
    )
    orig_im = tileHE.image
    m = t.F(orig_im)
    t.apply(tileHE)
    assert np.array_equal(tileHE.masks["testing"], m)


@pytest.mark.parametrize("transform", [LabelArtifactTileHE, LabelWhiteSpaceHE])
def test_binary_label_transforms(tileHE, transform):
    t = transform(label_name="test_label")
    t.apply(tileHE)
    assert tileHE.labels["test_label"] in [True, False]


def test_segment_mif(tileVectra):
    from pathml.preprocessing.transforms import SegmentMIFRemote

    vectra_collapse = CollapseRunsVectra()
    vectra_collapse.apply(tileVectra)
    t = SegmentMIFRemote(
        nuclear_channel=0,
        cytoplasm_channel=1,
        postprocess_kwargs_nuclear={
            "label_erosion": 10,
            "small_objects_threshold": 0.2,
            "fill_holes_threshold": 0.2,
            "pixel_expansion": 10,
            "maxima_algorithm": "peak_local_max",
        },
    )
    orig_im = tileVectra.image
    cell, nuclear = t.F(orig_im)
    t.apply(tileVectra)
    assert np.array_equal(tileVectra.masks["cell_segmentation"], cell)
    assert np.array_equal(tileVectra.masks["nuclear_segmentation"], nuclear)


def test_quantify_mif(tileVectra):
    from pathml.preprocessing.transforms import SegmentMIFRemote

    t = QuantifyMIF("cell_segmentation")
    with pytest.raises(AssertionError):
        t.apply(tileVectra)
    t2 = SegmentMIFRemote(nuclear_channel=0, cytoplasm_channel=1)
    vectra_collapse = CollapseRunsVectra()
    vectra_collapse.apply(tileVectra)
    t2.apply(tileVectra)
    t.apply(tileVectra)
    assert tileVectra.counts


def test_quantify_mif_coords():
    # make sure that QuantifyMIF uses i,j coordinates
    # make fake im with a single region
    im = np.zeros(shape=(50, 50, 3))
    seg = np.zeros(shape=(50, 50, 1), dtype=np.uint8)
    # add a region with known centroid at (i=27, j=7) == (x=7, y=27)
    seg[25:30, 5:10] = 1
    tile = Tile(image=im, coords=(0, 0), masks={"seg": seg}, stain="Fluor")
    quantifymif = QuantifyMIF(segmentation_mask="seg")
    quantifymif.apply(tile)
    assert tile.counts.obs.x[0] == 7
    assert tile.counts.obs.y[0] == 27


def test_collapse_runs_vectra(tileVectra):
    t = CollapseRunsVectra()
    orig_im = tileVectra.image
    m = t.F(orig_im)
    t.apply(tileVectra)
    assert np.array_equal(m, tileVectra.image)
    assert len(m.shape) == 3


@pytest.mark.parametrize(
    "transform",
    [
        MedianBlur(),
        GaussianBlur(),
        BoxBlur(),
        BinaryThreshold(),
        MorphOpen(),
        MorphClose(),
        ForegroundDetection(),
        SuperpixelInterpolation(),
        StainNormalizationHE(),
        NucleusDetectionHE(),
        TissueDetectionHE(),
        LabelArtifactTileHE(),
        LabelWhiteSpaceHE(),
        QuantifyMIF(segmentation_mask="test"),
        CollapseRunsVectra(),
        CollapseRunsCODEX(z=0),
        AdaptiveHistogramEqualization(),
        HistogramEqualization(),
        RescaleIntensity(),
    ],
)
def test_repr(transform):
    repr(transform)


def test_segmentMIF_repr():
    from pathml.preprocessing.transforms import SegmentMIFRemote

    repr(SegmentMIFRemote(nuclear_channel=0, cytoplasm_channel=1))


def test_collapse_runs_codex_repr():
    z_value = 2
    collapse_runs_codex = CollapseRunsCODEX(z=z_value)
    assert repr(collapse_runs_codex) == f"CollapseRunsCODEX(z={z_value})"


@pytest.mark.parametrize("z_value", [0, 1, 2])  # Example z-values
def test_collapse_runs_codex(tileCODEX, z_value):
    # Assuming tileCODEX is a pre-prepared CODEX tile for testing
    t = CollapseRunsCODEX(z=z_value)
    orig_im = tileCODEX.image

    # Manual calculation of expected result
    expected_image = orig_im[:, :, z_value, :]
    combined_channels = orig_im.shape[3] * orig_im.shape[4]
    expected_image = expected_image.reshape(
        orig_im.shape[0], orig_im.shape[1], combined_channels
    )

    # Apply transformation
    t.apply(tileCODEX)

    # Assert conditions
    assert np.array_equal(
        tileCODEX.image, expected_image
    ), "Image transformation did not match expected result"
    assert (
        len(tileCODEX.image.shape) == 3
    ), "Transformed image does not have correct shape"