Skip to content

RadHarmony

MIMIC-CXR (DICOM)

f10409/RadHarmony

MIMIC-CXR (DICOM)¶

Modality: CXR | Format: DICOM | Dim: 2D | Labels: 14 pathologies + reports

Overview¶

MIMIC-CXR is a large publicly available dataset of chest radiographs in DICOM format from the Beth Israel Deaconess Medical Center, containing 227,835 imaging studies for 65,379 patients (377,110 images total). It includes free-text radiology reports and CheXpert-extracted labels. Access requires credentialing on PhysioNet.

Download¶

Available at PhysioNet: MIMIC-CXR. Requires CITI training and signed DUA.

Expected layout¶

mimic-cxr/2.1.0/
  cxr-record-list.csv.gz
  cxr-study-list.csv.gz
  files/
    p10/
      p10000032/
        s50414267/
          02aa804e-bde0afdd-...dcm

Label columns¶

Same 14 labels as CheXpert (NLP-extracted from reports): atelectasis, cardiomegaly, consolidation, edema, enlarged_cardiomediastinum, fracture, lung_lesion, lung_opacity, no_finding, pleural_effusion, pleural_other, pneumonia, pneumothorax, support_devices

Constructor arguments¶

Argument	Type	Required	Default	Description
`base_image_dir`	`str`	Yes	`None`	`files/` subtree root (e.g. `/data/mimic-cxr/2.1.0/files/`) — direct parent of the `p10/`, `p11/`, … patient prefix dirs
`dicom_base_dir`	`str`	No	same as `base_image_dir`	Explicit DICOM root if different
`csv_path`	`str`	No	auto	`cxr-record-list.csv.gz`; auto-discovered
`label_csv_path`	`str`	No	auto	CheXpert labels CSV (from MIMIC-CXR-JPG)
`report_csv_path`	`str`	No	auto	`cxr-study-list.csv.gz` for report text
`drop_uncertain`	`bool`	No	`True`	Drop rows with uncertain labels

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 MIMICCXRDataset

ds = MIMICCXRDataset(
    base_image_dir="/data/mimic-cxr/2.1.0/files/",
    label_csv_path="/data/mimic-cxr-jpg/2.0.0/mimic-cxr-2.0.0-chexpert.csv",
    output_cls=True,
    output_report=True,
    dtype=torch.float32,
)
train_ds, val_ds = ds.get_datasets(n_splits=5)

Harmonizer¶

from radharmony.harmonizer import MIMICCXRHarmonizer

h = MIMICCXRHarmonizer(
    csv_path="/data/mimic-cxr/2.1.0/cxr-record-list.csv.gz",
    dicom_base_dir="/data/mimic-cxr/2.1.0/files/",
    label_csv_path="/data/mimic-cxr-jpg/2.0.0/mimic-cxr-2.0.0-chexpert.csv",
    report_csv_path="/data/mimic-cxr/2.1.0/cxr-study-list.csv.gz",
    report_base_dir="/data/mimic-cxr/2.1.0/files/",
)
df = h.harmonize()
print(df.columns.tolist())
df.to_csv("mimic_cxr_harmonized.csv", index=False)

Load from saved harmonized CSV¶

import pandas as pd
from radharmony.dataset import MIMICCXRDataset

ds = MIMICCXRDataset(
    base_image_dir="/data/mimic-cxr/2.1.0/files/",
    harmonized_df=pd.read_csv("mimic_cxr_harmonized.csv"),
    output_cls=True,
)

Harmonizer notes¶

Primary CSV is cxr-record-list.csv.gz (image–study mapping)
Labels come from a separate CheXpert label CSV (available in MIMIC-CXR-JPG)
Reports come from cxr-study-list.csv.gz
Uncertain labels handled as in CheXpert

Outputs¶

Flag	Key	Shape	Notes
`output_cls=True`	`"cls"`	`(14,)`	CheXpert-extracted labels
`output_report=True`	`"report"`	`str`	Free-text radiology report

Example paths¶

Role	Path
`base_image_dir`	`/path/to/MIMIC-CXR-V2-AWS/files/`
`csv_path` (`mimic-cxr-2.0.0-metadata.csv`)	`/path/to/MIMIC_CXR/physionet.org/files/mimic-cxr-jpg/2.0.0/mimic-cxr-2.0.0-metadata.csv`
`label_csv_path` (`mimic-cxr-2.0.0-chexpert.csv`)	`/path/to/MIMIC_CXR/physionet.org/files/mimic-cxr-jpg/2.0.0/mimic-cxr-2.0.0-chexpert.csv`
`report_csv_path` (`cxr-study-list.csv.gz`)	`/path/to/MIMIC-CXR-V2-AWS/cxr-study-list.csv.gz`

Metadata and chexpert CSVs live in the JPG release tree, not in the DICOM tree. Pass them explicitly — auto-discovery may not find them when the two releases are stored under different mounts.