bids-neuroimaging

Organize, validate, and query neuroimaging datasets in BIDS 1.8 format using pybids, mne-bids, and datalad; covers EEG, fMRI, and derivatives conventions.

インストールへ移動

リポジトリ: xjtulyc/awesome-rosetta-skills
ソースの最終更新活動: 2026年3月17日 09:21
検出された SKILL.md の言語: 英語
スター: 34
フォーク: 6

インストール方法

デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。

ソースファイルを確認

インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。

検索に戻るインストールに戻る

name	bids-neuroimaging
description	Organize, validate, and query neuroimaging datasets in BIDS 1.8 format using pybids, mne-bids, and datalad; covers EEG, fMRI, and derivatives conventions.
tags	["neuroscience","bids","neuroimaging","mne-bids","pybids","data-management"]
version	1.0.0
authors	[{"name":"awesome-rosetta-skills contributors","github":"@xjtulyc"}]
license	MIT
platforms	["claude-code","codex","gemini-cli","cursor"]
dependencies	{"python":["mne>=1.5.0","mne-bids>=0.13.0","pybids>=0.16.0","pandas>=2.0.0","numpy>=1.24.0","datalad>=0.19.0"],"system":["nodejs>=18.0.0"]}
last_updated	2026-03-17
status	stable

BIDS Neuroimaging: Organizing and Querying Brain Data

One-line summary: This Skill helps researchers convert raw neuroimaging data to BIDS 1.8 format, validate the structure, query datasets with pybids, and share via OpenNeuro/datalad.

When to Use This Skill

Use this Skill in the following scenarios:

When you need to organize EEG, MEG, or fMRI data into a standards-compliant BIDS directory
When your data has a multi-subject, multi-session structure requiring consistent naming
When you need to query a BIDS dataset to filter by task, subject, run, or modality
When you are using OpenNeuro to download or share public neuroimaging datasets
When you need to produce events.tsv files and sidecar JSON metadata from experimental logs
When preprocessing outputs must conform to BIDS derivatives conventions

Trigger keywords: BIDS, Brain Imaging Data Structure, pybids, BIDSLayout, mne-bids, write_raw_bids, events.tsv, participants.tsv, dataset_description.json, datalad, OpenNeuro, neuroimaging data organization, BIDS validator

Background & Key Concepts

BIDS Directory Hierarchy

The Brain Imaging Data Structure (BIDS; Gorgolewski et al., 2016) is a community standard for organizing neuroimaging data. The top-level layout is:

<dataset_root>/
  dataset_description.json   # Dataset-level metadata (required)
  participants.tsv            # Demographic table
  participants.json           # Column descriptions
  sub-<label>/
    ses-<label>/              # Optional session folder
      eeg/                    # Modality folder (eeg, meg, fmri, anat, ...)
        sub-<label>_ses-<label>_task-<label>_run-<index>_eeg.edf
        sub-<label>_ses-<label>_task-<label>_run-<index>_eeg.json  # Sidecar
        sub-<label>_ses-<label>_task-<label>_run-<index>_channels.tsv
        sub-<label>_ses-<label>_task-<label>_run-<index>_events.tsv
  derivatives/
    pipeline-name/
      sub-<label>/...         # Preprocessed outputs follow same layout

BIDS entities are key-value pairs separated by underscores. The order is fixed: sub → ses → task → acq → run → suffix.extension.

Events and Timing

The events.tsv file records stimulus/response onsets with three mandatory columns:

Column

Approach	Best for	Key assumption	Limitation
BIDS 1.8	Multi-modal human neuroimaging	Consistent entities across files	Learning curve for new modalities
NWB 2.x	Invasive electrophysiology	Hierarchical HDF5 objects	Less adopted for non-invasive EEG
EEGLab STUDY	EEG-only pipelines	MATLAB ecosystem	Not interoperable with Python tools
LORIS	Clinical database	Longitudinal patient data	Infrastructure-heavy deployment

# Create a virtual environment (recommended)
python -m venv .venv
source .venv/bin/activate   # Linux/macOS
# .venv\Scripts\activate    # Windows

# Install Python dependencies
pip install mne>=1.5.0 mne-bids>=0.13.0 pybids>=0.16.0 pandas>=2.0.0 numpy>=1.24.0

# Install datalad for dataset management
pip install datalad>=0.19.0

# Install the BIDS validator (requires Node.js >= 18)
npm install -g bids-validator

# Run this to confirm the environment is configured correctly
import mne
import mne_bids
import bids

print(f"mne version:       {mne.__version__}")
print(f"mne-bids version:  {mne_bids.__version__}")
print(f"pybids version:    {bids.__version__}")
# Expected: versions 1.5+, 0.13+, 0.16+ respectively

# Verify BIDS validator
bids-validator --version
# Expected: 1.x.x or higher

import json
import pandas as pd
from pathlib import Path

BIDS_ROOT = Path("my_bids_dataset")
BIDS_ROOT.mkdir(parents=True, exist_ok=True)

# --- dataset_description.json -------------------------------------------------
dataset_description = {
    "Name": "Auditory Oddball EEG Study",
    "BIDSVersion": "1.8.0",
    "DatasetType": "raw",
    "License": "CC0",
    "Authors": ["Jane Smith", "John Doe"],
    "Acknowledgements": "Data collected at ExampleLab.",
    "HowToAcknowledge": "Please cite Smith et al. (2024).",
    "ReferencesAndLinks": ["https://doi.org/10.xxxx/example"],
    "DatasetDOI": "",
}

with open(BIDS_ROOT / "dataset_description.json", "w") as f:
    json.dump(dataset_description, f, indent=2)
print("Written: dataset_description.json")

# --- participants.tsv ---------------------------------------------------------
participants = pd.DataFrame(
    {
        "participant_id": ["sub-01", "sub-02", "sub-03"],
        "age": [24, 31, 28],
        "sex": ["M", "F", "M"],
        "handedness": ["right", "right", "left"],
        "group": ["control", "control", "patient"],
    }
)
participants.to_csv(BIDS_ROOT / "participants.tsv", sep="\t", index=False)

# Companion sidecar describes each column
participants_json = {
    "age": {"Description": "Age at time of scan", "Units": "years"},
    "sex": {
        "Description": "Biological sex",
        "Levels": {"M": "male", "F": "female"},
    },
    "handedness": {
        "Description": "Dominant hand",
        "Levels": {"right": "right-handed", "left": "left-handed"},
    },
    "group": {
        "Description": "Study group",
        "Levels": {"control": "healthy control", "patient": "clinical patient"},
    },
}
with open(BIDS_ROOT / "participants.json", "w") as f:
    json.dump(participants_json, f, indent=2)
print("Written: participants.tsv and participants.json")

import mne
import mne_bids
import numpy as np
import pandas as pd
from pathlib import Path

BIDS_ROOT = Path("my_bids_dataset")

# --- Simulate a raw EEG file (replace with mne.io.read_raw_edf / read_raw_bdf) ---
n_channels = 32
sfreq = 250.0
duration_s = 120.0
n_samples = int(sfreq * duration_s)

rng = np.random.default_rng(seed=42)
data = rng.normal(scale=5e-6, size=(n_channels, n_samples))  # ~5 µV noise

ch_names = [
    "Fp1", "Fp2", "F7", "F3", "Fz", "F4", "F8",
    "T7", "C3", "Cz", "C4", "T8",
    "P7", "P3", "Pz", "P4", "P8",
    "O1", "Oz", "O2",
    "FC1", "FC2", "CP1", "CP2",
    "AF3", "AF4", "FC5", "FC6", "CP5", "CP6",
    "EOGh", "EOGv",
]
ch_types = ["eeg"] * 30 + ["eog", "eog"]

info = mne.create_info(ch_names=ch_names, sfreq=sfreq, ch_types=ch_types)
info["line_freq"] = 50.0  # European power line

raw = mne.io.RawArray(data, info)
montage = mne.channels.make_standard_montage("standard_1020")
raw.set_montage(montage, on_missing="ignore")

# --- Add stimulus annotations (simulated events at 250, 750, 1250 ... ms) ---
event_onsets = np.arange(1.0, 100.0, 2.5)  # Every 2.5 s
event_durations = np.zeros(len(event_onsets))
# Alternate standard (80%) and deviant (20%)
event_descriptions = [
    "deviant" if i % 5 == 4 else "standard"
    for i in range(len(event_onsets))
]
annotations = mne.Annotations(
    onset=event_onsets,
    duration=event_durations,
    description=event_descriptions,
)
raw.set_annotations(annotations)

# --- Define BIDSPath ---
bids_path = mne_bids.BIDSPath(
    subject="01",
    session="01",
    task="oddball",
    run="1",
    datatype="eeg",
    root=BIDS_ROOT,
)

# --- Write to BIDS (creates sidecar JSON, channels.tsv, events.tsv) ---
mne_bids.write_raw_bids(
    raw,
    bids_path=bids_path,
    overwrite=True,
    verbose=False,
    format="EDF",          # Save as EDF; use "auto" to keep original format
    events=None,           # Derived from annotations automatically
    event_id={"standard": 1, "deviant": 2},
)
print(f"BIDS EEG written to: {bids_path.fpath}")

# Inspect generated sidecar
sidecar_path = bids_path.copy().update(suffix="eeg", extension=".json").fpath
import json
with open(sidecar_path) as f:
    sidecar = json.load(f)
print("EEG sidecar JSON keys:", list(sidecar.keys()))

Parameter	Meaning	Recommended value	Notes
`format`	Output file format	`"EDF"` or `"auto"`	`"auto"` preserves original
`overwrite`	Overwrite existing files	`True` during development	Set to `False` for production
`event_id`	Annotation-to-code mapping	Dict matching annotation descriptions	Used to build events.tsv
`verbose`	Verbosity level	`False` or `"WARNING"`	MNE log level string or bool

from bids import BIDSLayout
import pandas as pd
from pathlib import Path

BIDS_ROOT = Path("my_bids_dataset")

# --- Initialize layout (indexes all files on first call) ---------------------
layout = BIDSLayout(str(BIDS_ROOT), validate=False)

# --- High-level queries -------------------------------------------------------
subjects = layout.get_subjects()
tasks    = layout.get_tasks()
sessions = layout.get_sessions()
print(f"Subjects: {subjects}")
print(f"Tasks:    {tasks}")
print(f"Sessions: {sessions}")

# --- Filter files by entity --------------------------------------------------
# Get all EEG data files for sub-01
eeg_files = layout.get(
    subject="01",
    datatype="eeg",
    suffix="eeg",
    extension=[".edf", ".bdf", ".fif", ".set"],
)
for f in eeg_files:
    print(f"  {f.path}")

# --- Access the events.tsv for a specific file --------------------------------
events_files = layout.get(
    subject="01",
    task="oddball",
    suffix="events",
    extension=".tsv",
)
for ef in events_files:
    events_df = pd.read_csv(ef.path, sep="\t")
    print(f"\nevents.tsv — {ef.path}")
    print(events_df.head())

# --- Build a summary DataFrame of all EEG files ------------------------------
all_eeg = layout.get(datatype="eeg", suffix="eeg", extension=".edf")
summary_records = []
for f in all_eeg:
    summary_records.append(
        {
            "subject":  f.entities.get("subject"),
            "session":  f.entities.get("session"),
            "task":     f.entities.get("task"),
            "run":      f.entities.get("run"),
            "path":     f.path,
        }
    )
summary_df = pd.DataFrame(summary_records)
print("\nDataset summary:")
print(summary_df.to_string())

# Validate the entire dataset (requires Node.js bids-validator)
bids-validator my_bids_dataset --verbose

# Non-interactive mode (suitable for CI pipelines)
bids-validator my_bids_dataset --json 2>/dev/null | python -c "
import sys, json
report = json.load(sys.stdin)
errors = report.get('issues', {}).get('errors', [])
warnings = report.get('issues', {}).get('warnings', [])
print(f'Errors: {len(errors)}, Warnings: {len(warnings)}')
for e in errors[:5]:
    print(' ERROR:', e.get('key'), '-', e.get('reason'))
"

import mne
import mne_bids
from pathlib import Path

BIDS_ROOT = Path("my_bids_dataset")
DERIV_ROOT = BIDS_ROOT / "derivatives" / "mne-preprocessing"
DERIV_ROOT.mkdir(parents=True, exist_ok=True)

# Write derivatives dataset_description.json
import json
deriv_desc = {
    "Name": "MNE Preprocessing Pipeline",
    "BIDSVersion": "1.8.0",
    "DatasetType": "derivative",
    "GeneratedBy": [
        {
            "Name": "mne-bids",
            "Version": mne_bids.__version__,
            "CodeURL": "https://mne.tools/mne-bids/",
        }
    ],
    "SourceDatasets": [
        {"DOI": "doi:10.xxxx/example", "URL": "https://openneuro.org/datasets/ds001"}
    ],
}
with open(DERIV_ROOT / "dataset_description.json", "w") as f:
    json.dump(deriv_desc, f, indent=2)

# Assume `raw_clean` is a preprocessed MNE Raw object
# Save preprocessed EEG as derivatives
deriv_path = mne_bids.BIDSPath(
    subject="01",
    session="01",
    task="oddball",
    run="1",
    datatype="eeg",
    suffix="eeg",
    description="preprocessed",   # desc-preprocessed entity
    root=DERIV_ROOT,
    check=False,                   # Skip BIDS validation for derivatives
)

# Write epochs as derivative (FIF format preserves MNE metadata)
# epochs.save(str(deriv_path.update(suffix="epo", extension=".fif").fpath), overwrite=True)
print(f"Derivatives path: {deriv_path.fpath}")

# Install datalad and the git-annex backend
pip install datalad
# On Linux: sudo apt install git-annex
# On macOS: brew install git-annex

# List available datasets (browse openneuro.org for ds-numbers)
# Install (clone metadata only; no large files downloaded yet)
datalad install https://github.com/OpenNeuroDatasets/ds003490.git
cd ds003490

# Download only sub-01 data
datalad get sub-01/

# Download the entire dataset (may be large)
datalad get .

import subprocess
from pathlib import Path

def datalad_get_subject(dataset_url: str, subject_id: str, local_dir: str) -> bool:
    """
    Clone an OpenNeuro dataset and download a single subject's files.

    Parameters
    ----------
    dataset_url : str
        GitHub URL of the OpenNeuro dataset, e.g.
        'https://github.com/OpenNeuroDatasets/ds003490.git'
    subject_id : str
        Subject label without 'sub-' prefix, e.g. '01'
    local_dir : str
        Local path to clone into.

    Returns
    -------
    bool : True if all commands succeeded.
    """
    local_path = Path(local_dir)
    if not local_path.exists():
        result = subprocess.run(
            ["datalad", "install", dataset_url, str(local_path)],
            capture_output=True, text=True,
        )
        if result.returncode != 0:
            print(f"datalad install failed: {result.stderr}")
            return False
        print(f"Installed dataset to {local_path}")

    subject_dir = local_path / f"sub-{subject_id}"
    result = subprocess.run(
        ["datalad", "get", str(subject_dir)],
        capture_output=True, text=True, cwd=str(local_path),
    )
    if result.returncode != 0:
        print(f"datalad get failed: {result.stderr}")
        return False

    print(f"Downloaded sub-{subject_id} to {subject_dir}")
    return True


# Example (replace with a real dataset URL)
# datalad_get_subject(
#     "https://github.com/OpenNeuroDatasets/ds003490.git",
#     subject_id="01",
#     local_dir="ds003490_local",
# )

import pandas as pd
import numpy as np
from pathlib import Path

def log_to_events_tsv(
    log_csv: str,
    onset_col: str,
    trial_type_col: str,
    duration_col: str | None = None,
    extra_cols: list[str] | None = None,
) -> pd.DataFrame:
    """
    Convert a stimulus presentation log CSV to BIDS events.tsv format.

    Parameters
    ----------
    log_csv : str
        Path to the experiment log file.
    onset_col : str
        Column containing event onset times in seconds.
    trial_type_col : str
        Column containing event type labels.
    duration_col : str, optional
        Column for event durations. If None, duration is set to 0.
    extra_cols : list of str, optional
        Additional columns to include (e.g., 'response_time', 'accuracy').

    Returns
    -------
    pd.DataFrame with BIDS-compliant column order.
    """
    log_df = pd.read_csv(log_csv)

    events = pd.DataFrame()
    events["onset"] = log_df[onset_col].astype(float)
    events["duration"] = (
        log_df[duration_col].astype(float) if duration_col else 0.0
    )
    events["trial_type"] = log_df[trial_type_col].astype(str)

    if extra_cols:
        for col in extra_cols:
            if col in log_df.columns:
                events[col] = log_df[col]

    events = events.sort_values("onset").reset_index(drop=True)
    return events


# Demonstration with synthetic log data
np.random.seed(0)
n_trials = 40
synthetic_log = pd.DataFrame(
    {
        "onset_s": np.sort(np.random.uniform(2.0, 100.0, n_trials)),
        "condition": np.random.choice(["standard", "deviant"], n_trials, p=[0.8, 0.2]),
        "rt_s": np.where(
            np.random.rand(n_trials) < 0.85,
            np.random.normal(0.45, 0.12, n_trials),
            np.nan,
        ),
        "correct": np.random.choice([0, 1], n_trials, p=[0.15, 0.85]),
    }
)
synthetic_log.to_csv("/tmp/example_log.csv", index=False)

events_df = log_to_events_tsv(
    log_csv="/tmp/example_log.csv",
    onset_col="onset_s",
    trial_type_col="condition",
    extra_cols=["rt_s", "correct"],
)
print(events_df.head(8).to_string())

# Save to BIDS location
output_tsv = Path("my_bids_dataset/sub-01/ses-01/eeg") / \
    "sub-01_ses-01_task-oddball_run-1_events.tsv"
output_tsv.parent.mkdir(parents=True, exist_ok=True)
events_df.to_csv(output_tsv, sep="\t", index=False)
print(f"Saved: {output_tsv}")

# Verify all three required columns exist
import pandas as pd
events = pd.read_csv("sub-01_task-oddball_events.tsv", sep="\t")
required = {"onset", "duration", "trial_type"}
missing = required - set(events.columns)
if missing:
    raise ValueError(f"Missing required events.tsv columns: {missing}")

from pathlib import Path
import json

bids_root = Path("my_bids_dataset")
desc_path = bids_root / "dataset_description.json"
if not desc_path.exists():
    json.dump(
        {"Name": "My Dataset", "BIDSVersion": "1.8.0"},
        open(desc_path, "w"),
        indent=2,
    )
    print("Created minimal dataset_description.json")

montage = mne.channels.make_standard_montage("standard_1020")
raw.set_montage(montage, on_missing="ignore")

Package	Tested versions	Known issues
mne	1.5, 1.6, 1.7	API changes in channel type handling between 1.5 and 1.6
mne-bids	0.13, 0.14	`write_raw_bids` format parameter changed in 0.13
pybids	0.16, 0.17	validate=True requires full BIDS layout; use False for partial datasets
datalad	0.19, 0.20	Requires git-annex system dependency; install separately

# =============================================
# End-to-end example: Multi-subject BIDS conversion
# Requirements: Python 3.10+; mne>=1.5.0, mne-bids>=0.13.0
# =============================================

import os
import mne
import mne_bids
import numpy as np
import json
from pathlib import Path

BIDS_ROOT = Path("example_bids_dataset")
BIDS_ROOT.mkdir(exist_ok=True)

# Write dataset_description.json
json.dump(
    {
        "Name": "Oddball Study",
        "BIDSVersion": "1.8.0",
        "DatasetType": "raw",
        "License": "CC0",
        "Authors": ["Example Researcher"],
    },
    open(BIDS_ROOT / "dataset_description.json", "w"),
    indent=2,
)

def make_synthetic_raw(seed: int) -> mne.io.BaseRaw:
    """Create a reproducible synthetic EEG Raw object."""
    rng = np.random.default_rng(seed)
    n_ch, sfreq, n_s = 19, 250.0, 30000
    ch_names = ["Fp1","Fp2","F7","F3","Fz","F4","F8","T7","C3","Cz",
                 "C4","T8","P7","P3","Pz","P4","P8","O1","O2"]
    info = mne.create_info(ch_names, sfreq=sfreq, ch_types="eeg")
    info["line_freq"] = 50.0
    data = rng.normal(scale=5e-6, size=(n_ch, n_s))
    raw = mne.io.RawArray(data, info)
    montage = mne.channels.make_standard_montage("standard_1020")
    raw.set_montage(montage, on_missing="ignore")
    # Add synthetic events
    onsets = np.arange(1.0, 100.0, 2.0)
    descs = ["deviant" if i % 5 == 4 else "standard" for i in range(len(onsets))]
    raw.set_annotations(mne.Annotations(onsets, np.zeros(len(onsets)), descs))
    return raw

# Convert 2 subjects x 2 sessions
for sub_idx, sub in enumerate(["01", "02"], start=1):
    for ses_idx, ses in enumerate(["01", "02"], start=1):
        seed_val = sub_idx * 100 + ses_idx
        raw = make_synthetic_raw(seed=seed_val)

        bp = mne_bids.BIDSPath(
            subject=sub, session=ses, task="oddball", run="1",
            datatype="eeg", root=BIDS_ROOT,
        )
        mne_bids.write_raw_bids(
            raw, bids_path=bp, overwrite=True, verbose=False,
            format="EDF",
            event_id={"standard": 1, "deviant": 2},
        )
        print(f"  Written: sub-{sub} ses-{ses}")

# Query the resulting layout
from bids import BIDSLayout
layout = BIDSLayout(str(BIDS_ROOT), validate=False)
print(f"\nSubjects: {layout.get_subjects()}")
print(f"Tasks:    {layout.get_tasks()}")
print(f"EEG files: {len(layout.get(suffix='eeg', extension='.edf'))}")
# Expected:
# Subjects: ['01', '02']
# Tasks: ['oddball']
# EEG files: 4

# =============================================
# End-to-end example 2: BIDSLayout query + analysis table
# Requirements: pybids>=0.16.0, pandas>=2.0.0
# =============================================

import pandas as pd
from bids import BIDSLayout
from pathlib import Path

# Replace with your local dataset path (after datalad install + get)
DATASET_PATH = os.environ.get("BIDS_DATASET_PATH", "example_bids_dataset")

layout = BIDSLayout(DATASET_PATH, validate=False)

# Load participants table
participants_tsv = Path(DATASET_PATH) / "participants.tsv"
if participants_tsv.exists():
    participants_df = pd.read_csv(participants_tsv, sep="\t")
else:
    participants_df = pd.DataFrame(columns=["participant_id"])

# Build file inventory
records = []
for eeg_file in layout.get(datatype="eeg", suffix="eeg"):
    sub = eeg_file.entities.get("subject", "")
    ses = eeg_file.entities.get("session", "")
    task = eeg_file.entities.get("task", "")
    run = eeg_file.entities.get("run", "")

    # Find matching events.tsv
    ev_files = layout.get(
        subject=sub, session=ses, task=task, run=run,
        suffix="events", extension=".tsv",
    )
    n_events = 0
    n_deviant = 0
    if ev_files:
        ev_df = pd.read_csv(ev_files[0].path, sep="\t")
        n_events = len(ev_df)
        if "trial_type" in ev_df.columns:
            n_deviant = (ev_df["trial_type"] == "deviant").sum()

    records.append(
        {
            "subject": f"sub-{sub}",
            "session": ses,
            "task": task,
            "run": run,
            "eeg_path": eeg_file.path,
            "n_events": n_events,
            "n_deviant": n_deviant,
        }
    )

inventory = pd.DataFrame(records)

# Merge with participants demographics
inventory = inventory.merge(
    participants_df.rename(columns={"participant_id": "subject"}),
    on="subject",
    how="left",
)

print("Analysis inventory:")
print(inventory.to_string())
print(f"\nTotal EEG files: {len(inventory)}")
print(f"Mean events per file: {inventory['n_events'].mean():.1f}")
# Expected output: a table merging file paths with demographics and event counts