def create_tilesummary(wsiPath,
is_wsi,
tilesFolderPath,
img_pil: PIL.Image.Image,
img_pil_filtered: PIL.Image.Image,
wsi_original_width: int,
wsi_original_height: int,
wsi_scaled_width: int,
wsi_scaled_height: int,
tile_height: int,
tile_width: int,
scale_factor: int,
tile_score_thresh: float,
tile_scoring_function,
tile_naming_func,
level: int,
best_level_for_downsample: int = 0) -> TileSummary:
"""
Args:
"""
np_img = util.pil_to_np_rgb(img_pil)
np_img_filtered = util.pil_to_np_rgb(img_pil_filtered)
tile_sum = score_tiles(np_img, np_img_filtered, wsiPath, is_wsi,
tilesFolderPath, tile_height, tile_width,
scale_factor, wsi_original_width,
wsi_original_height, wsi_scaled_width,
wsi_scaled_height, tile_score_thresh,
tile_scoring_function, tile_naming_func, level,
best_level_for_downsample)
return tile_sum
def show_wsi_with_marked_tiles(figsize: Tuple[int] = (10, 10),
scale_factor: int = 32,
tilesummary: TileSummary = None,
wsi_path: pathlib.Path = None,
df_tiles: pandas.DataFrame = None,
level: int = 0):
"""
Either provide a TileSummary object or wsi_path, df_tiles and level.
Loads a whole slide image, scales it down, converts it into a numpy array and shows it with a grid overlay for all tiles
that passed scoring to visualize which tiles e.g. "tiles.WsiOrROIToTilesMultithreaded" calculated as worthy to keep.
Arguments:
figsize: Size of the plotted matplotlib figure containing the image.
scale_factor: The larger, the faster this method works, but the plotted image has less resolution.
tilesummary: a TileSummary object of one wsi
wsi_path: Path to a whole-slide image
df_tiles: A pandas dataframe from e.g. "tiles.WsiOrROIToTilesMultithreaded" with spacial information about all tiles
level: The level that was specified in e.g. "tiles.WsiOrROIToTilesMultithreaded". 0 means highest magnification.
"""
if tilesummary != None:
wsi_pil, large_w, large_h, new_w, new_h, best_level_for_downsample = wsi_to_scaled_pil_image(
tilesummary.wsi_path,
scale_factor=tilesummary.scale_factor,
level=tilesummary.level)
wsi_np = util.pil_to_np_rgb(wsi_pil)
boxes = []
for tile in tilesummary.top_tiles():
box = np.array([
tile.get_x(),
tile.get_y(),
tile.get_width(),
tile.get_height()
]) / scale_factor
boxes.append(box)
show_np_with_bboxes(wsi_np, boxes, figsize)
else:
wsi_pil, large_w, large_h, new_w, new_h, best_level_for_downsample = wsi_to_scaled_pil_image(
wsi_path, scale_factor=scale_factor, level=level)
wsi_np = util.pil_to_np_rgb(wsi_pil)
boxes = []
for index, row in df_tiles.iterrows():
if row['wsi_path'] == wsi_path:
box = np.array([
row['x_upper_left'], row['y_upper_left'],
row['pixels_width'], row['pixels_height']
]) / scale_factor
boxes.append(box)
show_np_with_bboxes(wsi_np, boxes, figsize)
def tile_to_np_tile(tile):
"""
Convert tile information into the corresponding tile as a NumPy image read from the whole-slide image file.
Args:
tile: Tile object.
Return:
Tile as a NumPy image.
"""
pil_img = tile_to_pil_tile(tile)
np_img = util.pil_to_np_rgb(pil_img)
return np_img
def add_tile_stats_to_top_tile_summary(pil_img, tiles, z):
np_sum = util.pil_to_np_rgb(pil_img)
sum_r, sum_c, sum_ch = np_sum.shape
np_stats = np_tile_stat_img(tiles)
st_r, st_c, _ = np_stats.shape
combo_c = sum_c + st_c
combo_r = max(sum_r, st_r + z)
combo = np.zeros([combo_r, combo_c, sum_ch], dtype=np.uint8)
combo.fill(255)
combo[0:sum_r, 0:sum_c] = np_sum
combo[z:st_r + z, sum_c:sum_c + st_c] = np_stats
result = util.np_to_pil(combo)
return result
def open_image_np(filename):
"""
Open an image (*.jpg, *.png, etc) as an RGB NumPy array.
Args:
filename: Name of the image file.
returns:
A NumPy representing an RGB image.
"""
pil_img = open_image(filename)
np_img = util.pil_to_np_rgb(pil_img)
return np_img
def slide_to_scaled_np_image(slide_number):
"""
Convert a WSI training slide to a scaled-down NumPy image.
Args:
slide_number: The slide number.
Returns:
Tuple consisting of scaled-down NumPy image, original width, original height, new width, and new height.
"""
pil_img, large_w, large_h, new_w, new_h = slide_to_scaled_pil_image(
slide_number)
np_img = util.pil_to_np_rgb(pil_img)
return np_img, large_w, large_h, new_w, new_h
def tile_to_np_tile(tile, is_wsi: bool):
"""
Convert tile information into the corresponding tile as a NumPy image read from the whole-slide image file.
Args:
tile: Tile object.
is_wsi: if true, a WSI format like .ndpi is assumed, if false, a format like png is assumed (ROI)
Return:
Tile as a NumPy image.
"""
pil_img = tile_to_pil_tile(tile, is_wsi)
np_img = util.pil_to_np_rgb(pil_img)
return np_img
def np_text(text,
w_border=TILE_TEXT_W_BORDER,
h_border=TILE_TEXT_H_BORDER,
font_size=TILE_TEXT_SIZE,
text_color=TILE_TEXT_COLOR,
background=TILE_TEXT_BACKGROUND_COLOR):
"""
Obtain a NumPy array image representation of text.
Args:
text: The text to convert to an image.
w_border: Tile text width border (left and right).
h_border: Tile text height border (top and bottom).
font_size: Size of font.
text_color: Tile text color.
background: Tile background color.
Returns:
NumPy array representing the text.
"""
pil_img = pil_text(text, w_border, h_border, font_size, text_color,
background)
np_img = util.pil_to_np_rgb(pil_img)
return np_img