metadata

annotations_creators:
  - other
language:
  - en
language_creators:
  - found
  - expert-generated
license:
  - mit
multilinguality:
  - monolingual
pretty_name: RSNA-ATD2023
size_categories:
  - 10K<n<100K
source_datasets:
  - extended|other
tags: []
task_categories:
  - image-segmentation
task_ids:
  - semantic-segmentation

📁 Dataset

This dataset only comprised of 205 series of CT scans in .png file with raw images and raw mask.

Data source: Kaggle RSNA 2023 Abdominal Trauma Detection

🚀 Setup

pip install datasets

🤩 Feel the Magic

Load Dataset

from datasets import load_dataset

data = load_dataset('ziq/RSNA-ATD2023')
print(data)

DatasetDict({
    train: Dataset({
        features: ['patient_id', 'series_id', 'frame_id', 'image', 'mask'],
        num_rows: 70291
    })
})

Set Labels

labels = ["background", "liver", "spleen", "right_kidney", "left_kidney", "bowel"]

Train Test Split

data = data['train'].train_test_split(test_size=0.2)

train, test = data['train'], data['test']

# train[0]['patient_id']
# train[0]['image'] -> PIL Image
# train[0]['mask'] -> PIL Image

Get Segmentation Mask

ids = 3
image, mask = train[ids]['image'], train[ids]['mask']

mask = np.array(mask)

Write Plotting Function

from matplotlib.colors import ListedColormap, BoundaryNorm

colors = ['#02020e', '#520e6d', '#c13a50', '#f57d15', '#fac62c', '#f4f88e'] # inferno
bounds = range(0, len(colors) + 1)

# Define the boundaries for each class in the colormap
cmap, norm = ListedColormap(colors), BoundaryNorm(bounds, len(colors))

# Plot the segmentation mask with the custom colormap
def plot_mask(mask, alpha=1.0):
    _, ax = plt.subplots()
    cax = ax.imshow(mask, cmap=cmap, norm=norm, alpha=alpha)
    cbar = plt.colorbar(cax, cmap=cmap, norm=norm, boundaries=bounds, ticks=bounds)
    cbar.set_ticks([])
    _labels = [""] + labels
    for i in range(1, len(_labels)):
        cbar.ax.text(2, -0.5 + i, _labels[i], ha='left', color=colors[i - 1], fontsize=8)
    plt.axis('off')
    plt.show()
    
def plot(image, mask, cmap='gray'):
    plt.imshow(image * np.where(mask > 0,1,0), cmap=cmap)
    plt.axis('off')
    plt.show()

Plot

fig = plt.figure(figsize=(16,16))
ax1 = fig.add_subplot(131)
plt.axis('off')
ax1.imshow(image, cmap='gray')
ax2 = fig.add_subplot(132)
plt.axis('off')
ax2.imshow(mask, cmap='gray')
ax3 = fig.add_subplot(133)
ax3.imshow(image*np.where(mask>0,1,0), cmap='gray')
plt.axis('off')
plt.show()

plot(image, mask)

Custom Color

plot_mask(mask)

Plot only one class (e.g. liver)

liver, spleen, right_kidney, left_kidney, bowel = [(mask == i,1,0)[0] * i for i in range(1, len(labels))]
plot_mask(liver)