RSNA Pediatric Bone Age¶

Modality: Radiograph | Format: PNG | Dim: 2D | Labels: age in months (regression)

Overview¶

The RSNA Pediatric Bone Age Challenge (2017) dataset contains 12,611 hand radiographs from patients aged 1 month to 228 months. The task is to predict skeletal age from the hand X-ray. This is a regression dataset — the target is age_months, not a classification label.

Download¶

Available from the RSNA website.

Expected layout — train and val are separate downloads and each takes its own base_image_dir. They do not have to share a parent directory.

Expected layout¶

Train (one ZIP):

<train base_image_dir>/                 # e.g. boneage-training-dataset/
  <id>.png                              # 12,611 PNGs
<train base_image_dir>/../train.csv     # sibling — auto-discovered

Val (separate ZIP):

<val base_image_dir>/                   # e.g. Bone Age Validation Set/
  Validation Dataset.csv
  boneage-validation-dataset-1/
    <id>.png                            # 800 PNGs (ids 1386–9708)
  boneage-validation-dataset-2/
    <id>.png                            # 625 PNGs (ids 10018–15612)

Label columns¶

This dataset uses REG_COLS instead of LABEL_COLS:

Column	Description
`age_months`	Skeletal age in months

Constructor arguments¶

Argument	Type	Required	Default	Description
`base_image_dir`	`str`	Yes	`None`	Split-specific image directory. Train: `boneage-training-dataset/` (flat dir of `<id>.png` files). Val: `Bone Age Validation Set/` (contains `Validation Dataset.csv` plus the two `boneage-validation-dataset-{1,2}/` sub-subdirs — val is a structural exception with multi-sibling sub-subdirs). Train and val are separate downloads and may live anywhere on disk
`csv_path`	`str`	No	auto	`train.csv` or `Validation Dataset.csv`; auto-discovered
`split`	`str`	No	`'train'`	Which split to load

Shared arguments (inherited from `BaseRadiologicalDataset`)¶

Argument	Type	Required	Default	Description
`output_cls`	`bool`	No	`False`	Include `"cls"` in data dict (no label cols — unused)
`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
`output_reg`	`bool`	No	`False`	Include `"reg"` tensor with age in months
`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 both RSNABoneAgeTrainDataset and RSNABoneAgeValDataset (both omit the split= arg — split is fixed by the class).

Train split¶

import torch
from radharmony.dataset import RSNABoneAgeTrainDataset

ds_train = RSNABoneAgeTrainDataset(
    base_image_dir="/data/boneage-training-dataset/",
    output_reg=True,
    dtype=torch.float32,
)
train_ds, _ = ds_train.get_datasets(n_splits=5)

Val split (separate download — different `base_image_dir`)¶

from radharmony.dataset import RSNABoneAgeValDataset

ds_val = RSNABoneAgeValDataset(
    base_image_dir="/data/Bone Age Validation Set/",
    output_reg=True,
    dtype=torch.float32,
)

Harmonizer¶

from radharmony.harmonizer import RSNABoneAgeTrainHarmonizer

h = RSNABoneAgeTrainHarmonizer(
    base_image_dir="/data/boneage-training-dataset/",
)
df = h.harmonize()
print(df.columns.tolist())
# ['patient_id', 'study_id', 'image_path', 'age_months']
df.to_csv("rsna_bone_age_harmonized.csv", index=False)

RSNABoneAgeValHarmonizer is the val-split equivalent.

Load from saved harmonized CSV¶

import pandas as pd
from radharmony.dataset import RSNABoneAgeTrainDataset

ds = RSNABoneAgeTrainDataset(
    base_image_dir="/data/boneage-training-dataset/",
    harmonized_df=pd.read_csv("rsna_bone_age_harmonized.csv"),
    output_reg=True,
)

Harmonizer notes¶

Train and val splits use different CSV files and different image directories
Use RSNABoneAgeTrainDataset for the train split and RSNABoneAgeValDataset for the val split — split is fixed by the class
REG_COLS = ["age_months"]; use output_reg=True to get the regression target
Age is in months (integer); model outputs should be in the same unit

Outputs¶

Flag	Key	Shape	Notes
`output_reg=True`	`"reg"`	`(1,)`	Skeletal age in months

Example paths¶

Role	Path
`base_image_dir` (train)	`/path/to/rsna-boneage/boneage-training-dataset/`
`base_image_dir` (val)	`/path/to/rsna-boneage/Bone Age Validation Set/`
`csv_path` (train, `train.csv`)	`/path/to/rsna-boneage/train.csv`
`csv_path` (val, `Validation Dataset.csv`)	`/path/to/rsna-boneage/Bone Age Validation Set/Validation Dataset.csv`