Skip to content

RadHarmony

CT-RATE ^beta

f10409/RadHarmony

CT-RATE¶

Modality: CT | Format: NIfTI | Dim: 3D | Labels: 18 findings

Overview¶

CT-RATE is a large-scale chest CT dataset containing 50,188 non-contrast chest CT volumes from 21,304 unique patients, paired with radiology reports and NLP-extracted labels for 18 pathological findings. It is one of the largest publicly available chest CT datasets with structured labels.

Download¶

Available on HuggingFace: ibrahimhamamci/CT-RATE. Requires HuggingFace account.

Expected layout¶

CT-RATE/dataset/
  train_fixed/
    train_predicted_labels.csv
    train_metadata.csv
    <volume_id>/
      <volume_id>.nii.gz

Label columns¶

Column	Description
`arterial_wall_calcification`	Arterial wall calcification
`atelectasis`	Atelectasis
`bronchiectasis`	Bronchiectasis
`cardiomegaly`	Cardiomegaly
`consolidation`	Consolidation
`coronary_artery_wall_calcification`	Coronary artery calcification
`emphysema`	Emphysema
`hiatal_hernia`	Hiatal hernia
`interlobular_septal_thickening`	Interlobular septal thickening
`lung_nodule`	Lung nodule
`lung_opacity`	Lung opacity
`lymphadenopathy`	Lymphadenopathy
`medical_material`	Medical material / device
`mosaic_attenuation_pattern`	Mosaic attenuation pattern
`peribronchial_thickening`	Peribronchial thickening
`pericardial_effusion`	Pericardial effusion
`pleural_effusion`	Pleural effusion
`pulmonary_fibrotic_sequela`	Pulmonary fibrotic sequela

Constructor arguments¶

Argument	Type	Required	Default	Description
`base_image_dir`	`str`	Yes	`None`	`train_fixed/` (or `validation_fixed/`) — root containing the two-level volume hierarchy `<train_X>/<train_X_Y>/<volume>.nii.gz` (e.g. `/data/CT-RATE/dataset/train_fixed/`)
`csv_path`	`str`	No	auto	`train_predicted_labels.csv`; auto-discovered
`view_position_csv_path`	`str`	No	auto	`train_metadata.csv`; auto-discovered
`hu_window`	`tuple[float,float]\\|None`	No	`(-1000, 1000)`	HU clip range

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¶

import torch
from radharmony.dataset import CTRATEDataset

ds = CTRATEDataset(
    base_image_dir="/data/CT-RATE/dataset/train_fixed/",
    output_cls=True,
    hu_window=(-1000, 400),   # lung window
    dtype=torch.float32,
)
train_ds, val_ds = ds.get_datasets(n_splits=5)

Harmonizer¶

from radharmony.harmonizer import CTRATEHarmonizer

h = CTRATEHarmonizer(
    csv_path="/data/CT-RATE/dataset/train_fixed/train_predicted_labels.csv",
    base_image_dir="/data/CT-RATE/dataset/train_fixed/",
    view_position_csv_path="/data/CT-RATE/dataset/train_fixed/train_metadata.csv",
)
df = h.harmonize()
print(df.columns.tolist())
df.to_csv("ct_rate_harmonized.csv", index=False)

Load from saved harmonized CSV¶

import pandas as pd
from radharmony.dataset import CTRATEDataset

ds = CTRATEDataset(
    base_image_dir="/data/CT-RATE/dataset/train_fixed/",
    harmonized_df=pd.read_csv("ct_rate_harmonized.csv"),
    output_cls=True,
)

Harmonizer notes¶

Labels are NLP-extracted from radiology reports; binary (1 = present, 0 = absent)
Metadata CSV provides view position and patient demographics
Images are NIfTI (.nii.gz); MONAI loads them natively

Outputs¶

Flag	Key	Shape	Notes
`output_cls=True`	`"cls"`	`(18,)`	Multi-label binary findings

Example paths¶

Role	Path
`base_image_dir` (train)	`/path/to/CT-RATE/dataset/train_fixed/`
`base_image_dir` (valid)	`/path/to/CT-RATE/dataset/valid_fixed/`
`csv_path` (train, `train_predicted_labels.csv`)	`/path/to/CT-RATE/dataset/tables/train_predicted_labels.csv`
`csv_path` (valid, `valid_predicted_labels.csv`)	`/path/to/CT-RATE/dataset/tables/valid_predicted_labels.csv`
`view_position_csv_path` (train, `train_metadata.csv`)	`/path/to/CT-RATE/dataset/tables/train_metadata.csv`
`view_position_csv_path` (valid, `validation_metadata.csv`)	`/path/to/CT-RATE/dataset/tables/validation_metadata.csv`

CSVs live in tables/ (sibling of train_fixed/ and valid_fixed/), auto-discoverable via infer_path.