def visualise_cut(slide,
list_pos,
res_to_view=None,
color='red',
size=12,
title=""):
map_col = {0: "white", 1: "red", 2: "green", 3: "blue"}
if res_to_view is None:
res_to_view = slide.level_count - 3
whole_slide = np.array(
slide.read_region((0, 0), res_to_view,
slide.level_dimensions[res_to_view]))
max_x, max_y = slide.level_dimensions[res_to_view]
fig = plt.figure(figsize=(size, size))
ax = fig.add_subplot(111, aspect='equal')
# ax.imshow(flip_vertical(whole_slide)) # , origin='lower')
# whole_slide = flip_horizontal(whole_slide)
ax.imshow(whole_slide)
for para in list_pos:
top_left_x, top_left_y = UOS.get_X_Y_from_0(slide, para[0], para[1],
res_to_view)
w, h = UOS.get_size(slide, para[2], para[3], para[4], res_to_view)
p = patches.Rectangle((top_left_x, max_y - top_left_y - h),
w,
h,
fill=False,
edgecolor=map_col[para[5]])
p = patches.Rectangle((top_left_x, top_left_y),
w,
h,
fill=False,
edgecolor=map_col[para[5]])
ax.add_patch(p)
ax.set_title(title, size=20)
return fig
whole_img = np.array(whole_img)[:, :, 0:3]
small_img = resize(whole_img, (500, 500))
small_img = img_as_ubyte(small_img)
mask = ROI_binary_mask(small_img, ticket=(80, 80, 80))
mask = mask.astype('uint8')
mask[mask > 0] = 255
mask_tissue = resize(mask, last_dim_inv, order=0)
mask_tissue = img_as_ubyte(mask_tissue)
mask_tissue[mask_tissue > 0] = 255
class_0 = np.zeros_like(whole_img[:, :, 0], dtype=float)
class_1 = np.zeros_like(whole_img[:, :, 0], dtype=float)
class_2 = np.zeros_like(whole_img[:, :, 0], dtype=float)
class_3 = np.zeros_like(whole_img[:, :, 0], dtype=float)
s_0_x, s_0_y = UOS.get_size(WSI, 2 * 448, 2 * 448, 0, last_dim_n)
model = load_model(WEIGHTS)
for x in range(0, whole_img.shape[0], s_0_x):
for y in range(0, whole_img.shape[1], s_0_y):
if mask_tissue[x, y] != 0:
x_0, y_0 = UOS.get_X_Y(WSI, y, x, last_dim_n)
img = resize(
np.array(WSI.read_region((x_0, y_0), 0,
(448 * 2, 448 * 2)))[:, :, 0:3],
(224, 224))
img = img_as_ubyte(img).astype(float) - mean
img = np.expand_dims(img, axis=0)
res = model.predict(img)
class_0[x:(x + s_0_x), y:(y + s_0_y)] = res[0][0]
class_1[x:(x + s_0_x), y:(y + s_0_y)] = res[0][1]
mask[mask > 0] = 255
mask_tissue = resize(mask, last_dim_inv, order=0)
mask_tissue = img_as_ubyte(mask_tissue)
mask_tissue[mask_tissue > 0] = 255
GT_file = imread(GT_path)
GT_file[GT_file > 0] = 255
Red_area = GT_file[:, :, 0]
Red_area = img_as_ubyte(resize(Red_area, last_dim_inv, order=0))
Green_area = GT_file[:, :, 1]
Green_area = img_as_ubyte(resize(Green_area, last_dim_inv, order=0))
Blue_area = GT_file[:, :, 2]
Blue_area = img_as_ubyte(resize(Blue_area, last_dim_inv, order=0))
mask_tissue = mask_tissue - Red_area - Green_area - Blue_area
s_0_x, s_0_y = UOS.get_size(scan, 448, 448, 0, last_dim_n)
channels = (mask_tissue, Red_area, Green_area, Blue_area)
list_channel = []
for k, c in enumerate(channels):
i = 0
c_cop = c.copy()
while ((c_cop > 0).sum() > 0 and i < number_of_imgs):
x_l, y_l = np.where(c > 0)
pic = random.randint(0, len(x_l)) - 1
x = x_l[pic] - s_0_x // 2
y = y_l[pic] - s_0_y // 2
c_cop[(x - s_0_x):(x + s_0_x), (y - s_0_y):(y + s_0_y)] = 0
x_0, y_0 = UOS.get_X_Y(scan, x, y, last_dim_n)
# add cube to final list: label, x,y for read_imd
small_img = resize(whole_img, (500, 500))
small_img = img_as_ubyte(small_img)
mask = ROI_binary_mask(small_img, ticket=(80,80,80))
mask = mask.astype('uint8')
mask[mask > 0] = 255
mask_tissue = resize(mask, last_dim_inv, order=0)
mask_tissue = img_as_ubyte(mask_tissue)
mask_tissue[mask_tissue > 0] = 255
class_0 = np.zeros_like(whole_img[:,:,0])
class_1 = np.zeros_like(whole_img[:,:,0])
class_2 = np.zeros_like(whole_img[:,:,0])
class_3 = np.zeros_like(whole_img[:,:,0])
s_0_x, s_0_y = UOS.get_size(WSI, 224, 224, 0, last_dim_n)
model = load_model(WEIGHTS)
for x in range(0, whole_img.shape[0], s_0_x):
for y in range(0, whole_img.shape[1], s_0_y):
if mask_tissue[x, y] != 0:
img = WSI.read_region((x-s_0_x/2, y-s_0_y/2), 0, (224, 224))
img = img.astype(float) - mean
img = np.expand_dims(img, axis=0)
res = model.predict(img)
class_0[x, y] = res[0]
class_1[x, y] = res[1]
class_2[x, y] = res[2]
class_3[x, y] = res[3]