SIIM COVID-19 Detection¶

Modality: CXR | Format: DICOM | Dim: 2D | Labels: 4 appearance classes + bboxes

Overview¶

The SIIM-FISABIO-RSNA COVID-19 Detection dataset from the 2021 Kaggle challenge contains 6,334 chest X-rays labelled for COVID-19 lung opacity appearance. Each study has a study-level label and images may have bounding box annotations for opacities. The dataset uses two-level annotations: study-level labels and image-level bounding boxes.

Download¶

Available on Kaggle: SIIM COVID-19 Detection. Requires Kaggle account.

Expected layout¶

siim-covid19-detection/
  train/
    train_study_level.csv
    train_image_level.csv
    <studyInstanceUID>/
      <seriesInstanceUID>/
        <sopInstanceUID>.dcm

Label columns¶

Column	Description
`atypical_appearance`	COVID-19 atypical appearance
`indeterminate_appearance`	Indeterminate appearance
`negative_for_pneumonia`	No pneumonia findings
`typical_appearance`	COVID-19 typical appearance

Constructor arguments¶

Argument	Type	Required	Default	Description
`base_image_dir`	`str`	Yes	`None`	`train/` (or `test/` for the test split) — root walked for `<study>/<series>/<sop>.dcm` files (e.g. `~/datasets/competitions/siim-covid19-detection/train/`)
`csv_path`	`str`	No	auto	`train_study_level.csv`; auto-discovered
`image_csv_path`	`str`	No	auto	`train_image_level.csv` for bbox annotations

Shared arguments (inherited from `BaseRadiologicalDataset`)¶

Argument	Type	Required	Default	Description
`output_cls`	`bool`	No	`False`	Include `"cls"` in data dict
`output_mask`	`bool`	No	`False`	Include `"mask"` in data dict
`output_report`	`bool`	No	`False`	Include `"report"` in data dict
`output_bbox`	`bool`	No	`False`	Include `"bbox"` and `"bbox_labels"` in data dict
`transform`	MONAI transform	No	`None`	MONAI Compose transform; `None` uses the default pipeline
`cache_dir`	`str`	No	`"./cache"`	MONAI cache directory
`dtype`	`torch.dtype`	No	`torch.bfloat16`	Output tensor dtype
`harmonized_df`	`pd.DataFrame`	No	`None`	Pre-built harmonized DataFrame
`harmonizer`	harmonizer	No	`None`	Pre-instantiated harmonizer
`harmonizer_path`	`str`	No	`None`	Path to saved harmonized CSV

Dataset constructor¶

The constructor arguments above apply to SIIMCOVID19TrainDataset. SIIMCOVID19TestDataset accepts the same args except csv_path and image_csv_path (the test split has no annotation CSVs — those are hardcoded to None internally). The test split also silently ignores output_cls and output_bbox.

Train split¶

import torch
from radharmony.dataset import SIIMCOVID19TrainDataset

ds = SIIMCOVID19TrainDataset(
    base_image_dir="/data/siim-covid19-detection/train/",
    output_cls=True,
    output_bbox=True,
    dtype=torch.float32,
)
train_ds, val_ds = ds.get_datasets(n_splits=5)

Test split¶

from radharmony.dataset import SIIMCOVID19TestDataset

ds = SIIMCOVID19TestDataset(
    base_image_dir="/data/siim-covid19-detection/test/",
    dtype=torch.float32,
)

Harmonizer¶

from radharmony.harmonizer import SIIMCOVID19TrainHarmonizer

h = SIIMCOVID19TrainHarmonizer(
    csv_path="/data/siim-covid19-detection/train_study_level.csv",
    base_image_dir="/data/siim-covid19-detection/train/",
    image_csv_path="/data/siim-covid19-detection/train_image_level.csv",
)
df = h.harmonize()
print(df.columns.tolist())
df.to_csv("siim_covid19_harmonized.csv", index=False)

Load from saved harmonized CSV¶

import pandas as pd
from radharmony.dataset import SIIMCOVID19TrainDataset

ds = SIIMCOVID19TrainDataset(
    base_image_dir="/data/siim-covid19-detection/train/",
    harmonized_df=pd.read_csv("siim_covid19_harmonized.csv"),
    output_cls=True,
)

Harmonizer notes¶

Study-level CSV provides the appearance class labels (one per study)
Image-level CSV provides bounding boxes (opacity locations, one per box instance)
A study may have multiple images; each image inherits the study-level label
Bounding boxes are normalised to fractional coordinates

Outputs¶

Flag	Key	Shape	Notes
`output_cls=True`	`"cls"`	`(4,)`	One-hot appearance class
`output_bbox=True`	`"bbox"`, `"bbox_labels"`	list	Opacity bounding boxes

Example paths¶

Role	Path
`base_image_dir` (train)	`/path/to/siim-covid19-detection/train/`
`base_image_dir` (test)	`/path/to/siim-covid19-detection/test/`
`csv_path` (`train_study_level.csv`)	`/path/to/siim-covid19-detection/train_study_level.csv`
`image_csv_path` (`train_image_level.csv`)	`/path/to/siim-covid19-detection/train_image_level.csv`

CSVs live at the competition root (parent of train/), auto-discoverable via infer_path. SIIMCOVID19TestDataset does NOT accept csv_path or image_csv_path.