Hagen dataset¶

data_description = (
    "Laser scanning fluorescence microscopy imaging of actin in bovine pulmonary "
    "artery endothelial cells, acquired with a low signal-to-noise ratio. This dataset "
    "is a subset of Dataset 4, as described in the paper by Hagen, G.M. et al., 2021 "
    "(Fluorescence Microscopy Datasets for Training Deep Neural Networks, "
    "GigaScience). The data is available on Zenodo at "
    "https://zenodo.org/records/10925855."
)
print(data_description)

# load training and validation image and show them side by side
train_image = tifffile.imread(file)
print(f"Image shape: {train_image.shape}")

fig, ax = plt.subplots(1, 3, figsize=(15, 5))
ax[0].imshow(
    train_image[0], 
    cmap="gray", 
    vmin=np.percentile(train_image[0], 1), 
    vmax=np.percentile(train_image[0], 99),
)
ax[0].set_title("Image 1")
ax[1].imshow(
    train_image[1], 
    cmap="gray", 
    vmin=np.percentile(train_image[1], 1), 
    vmax=np.percentile(train_image[1], 99),
)
ax[1].set_title("Image 2")
ax[2].imshow(
    train_image[2], 
    cmap="gray", 
    vmin=np.percentile(train_image[2], 1), 
    vmax=np.percentile(train_image[2], 99),
)
ax[2].set_title("Image 3")

autocorr = autocorrelation(train_image[0])

# crop the correlation around (0, 0)
midpoint = train_image[0].shape[0] // 2
crop_size = 20
slices = (
    slice(midpoint - crop_size, midpoint + crop_size),
    slice(midpoint - crop_size, midpoint + crop_size),
)
# plot autocorrelation
fig, ax = plt.subplots(1, 2, figsize=(15, 5))
ax[0].imshow(train_image[0], cmap="gray")
ax[0].set_title("Image 1")
ax[1].imshow(autocorr[slices], cmap="gray")
ax[1].set_title("Autocorrelation")

# create configuration
algo = "n2v2" if use_n2v2 else "n2v"

config = create_n2v_configuration(
    experiment_name="hagen_" + algo,
    data_type="array",
    axes="SYX",
    patch_size=(64, 64),
    batch_size=32,
    num_epochs=10,
    use_n2v2=use_n2v2,
)

# change augmentations
config.data_config.transforms[0].flip_y = False  # do not flip y
config.data_config.transforms.pop(1)  # remove 90 degree rotations

print(config)

n = 5

fig, ax = plt.subplots(n, 2, figsize=(10, n * 5))
for i in range(n):
    ax[i, 0].imshow(
        train_image[i], 
        cmap="gray", 
        vmin=np.percentile(train_image[i], 1), 
        vmax=np.percentile(train_image[i], 99),
    )
    ax[i, 0].set_title(f"Noisy - Image {i}")
    ax[i, 1].imshow(
        prediction[i].squeeze(), 
        cmap="gray", 
        vmin=np.percentile(prediction[i], 1), 
        vmax=np.percentile(prediction[i], 99),
    )
    ax[i, 1].set_title(f"Prediction - Prediction {i}")

# create a cover image
im_idx = 0
cv_image_noisy = train_image[im_idx]
cv_image_pred = prediction[im_idx].squeeze()

# create image
cover = np.zeros((256, 256))   
(height, width) = cv_image_noisy.shape
assert height > 256
assert width > 256

# normalize train and prediction
norm_noise = (cv_image_noisy - cv_image_noisy.min()) / (cv_image_noisy.max() - cv_image_noisy.min())
norm_pred = (cv_image_pred - cv_image_pred.min()) / (cv_image_pred.max() - cv_image_pred.min())

# fill in halves
cover[:, :256 // 2] = norm_noise[height // 2 - 256 // 2:height // 2 + 256 // 2, width // 2 - 256 // 2:width // 2]
cover[:, 256 // 2:] = norm_pred[height // 2 - 256 // 2:height // 2 + 256 // 2, width // 2:width // 2 + 256 // 2]

# plot the single image
plt.imshow(cover, cmap="gray")

# save the image
im = Image.fromarray(cover * 255)
im = im.convert('L')
im.save("Hagen_N2V.jpeg")

general_description = (
    "This model is a UNet trained with the Noise2Void algorithm to denoise images. The "
    "training data consists of laser scanning confocal microscopy images of actin in "
    "bovine pulmonary artery endothelial cells. The notebook used to train this model "
    "is available on the CAREamics documentation website at the following link: "
    "https://careamics.github.io/0.1/applications/Noise2Void/Hagen/."
)
print(general_description)