GLM Data

Statistical analysis results from General Linear Model (GLM) analyses of the LAION-fMRI dataset.

Overview

The GLM derivatives contain first-level (subject-level) and group-level statistical maps derived from analyzing the functional data.

File Organization

GLM results are stored in the derivatives folder:

derivatives/
└── glm/
    ├── first_level/
    │   └── sub-01/
    │       ├── sub-01_task-experiment_contrast-faceVsBaseline_stat-t.nii.gz
    │       ├── sub-01_task-experiment_contrast-faceVsBaseline_stat-z.nii.gz
    │       ├── sub-01_task-experiment_contrast-faceVsScene_stat-t.nii.gz
    │       ├── sub-01_task-experiment_contrast-faceVsScene_stat-z.nii.gz
    │       ├── sub-01_task-experiment_contrast-sceneVsBaseline_stat-t.nii.gz
    │       ├── sub-01_task-experiment_contrast-sceneVsBaseline_stat-z.nii.gz
    │       └── sub-01_task-experiment_design-matrix.png
    └── group_level/
        ├── group_contrast-faceVsBaseline_stat-t.nii.gz
        ├── group_contrast-faceVsBaseline_stat-z.nii.gz
        ├── group_contrast-faceVsBaseline_fdr-corrected.nii.gz
        ├── group_contrast-faceVsScene_stat-t.nii.gz
        ├── group_contrast-faceVsScene_stat-z.nii.gz
        └── group_contrast-faceVsScene_fdr-corrected.nii.gz

First-Level Results

Subject-Level Maps

Individual subject statistical maps for each contrast:

File Naming Convention:

sub-<label>_task-<label>_contrast-<label>_stat-<type>.nii.gz

Where:

• sub-<label>: Subject identifier (e.g., sub-01)

• task-<label>: Task name (e.g., task-experiment)

• contrast-<label>: Contrast name (e.g., faceVsBaseline)

• stat-<type>: Statistic type (t, z, effect)

Available Contrasts

Standard contrasts computed for each subject:

Single Category Contrasts:

• faceVsBaseline: Face stimuli vs. baseline

• sceneVsBaseline: Scene stimuli vs. baseline

• objectVsBaseline: Object stimuli vs. baseline

Differential Contrasts:

• faceVsScene: Faces vs. scenes

• faceVsObject: Faces vs. objects

• sceneVsObject: Scenes vs. objects

Design Matrix Visualization

Each subject includes a design matrix visualization:

sub-01_task-experiment_design-matrix.png

This shows:

• Task regressors (experimental conditions)

• Confound regressors (motion, signals)

• Temporal structure of the experiment

Loading First-Level Data

Using the Python package:

from laion_fmri_dataloader import LAIONfMRIDataset
from nilearn import plotting
import matplotlib.pyplot as plt

# Initialize dataset
dataset = LAIONfMRIDataset(data_dir='./laion-fmri-data')

# Load first-level contrast map
contrast_img = dataset.load_first_level_contrast(
    subject='sub-01',
    task='experiment',
    contrast='faceVsBaseline',
    stat_type='z'
)

# Display the statistical map
plotting.plot_stat_map(
    contrast_img,
    threshold=3.1,  # p < 0.001 uncorrected
    title='Face > Baseline (sub-01)',
    display_mode='ortho',
    cut_coords=(0, 0, 0)
)
plt.show()

# Load all contrasts for a subject
all_contrasts = dataset.load_all_contrasts(
    subject='sub-01',
    task='experiment',
    stat_type='z'
)

print(f"Available contrasts: {list(all_contrasts.keys())}")

Group-Level Results

Group Statistical Maps

Second-level analyses combining data across subjects:

File Naming Convention:

group_contrast-<label>_stat-<type>.nii.gz
group_contrast-<label>_<correction>-corrected.nii.gz

Where:

• contrast-<label>: Contrast name

• stat-<type>: Statistic type (t, z)

• <correction>: Multiple comparison correction method (fdr, fwe)

Multiple Comparisons Correction

Statistical maps are provided with different correction methods:

Uncorrected:

• Raw t-statistics and z-scores

• Useful for ROI analyses or hypothesis-driven research

FDR Corrected:

• False Discovery Rate correction

• Controls proportion of false positives

• More lenient than FWE

FWE Corrected:

• Family-Wise Error correction (when available)

• Controls probability of any false positive

• More conservative

Loading Group-Level Data

# Load group-level contrast
group_contrast = dataset.load_group_level_contrast(
    contrast='faceVsBaseline',
    stat_type='z'
)

# Load FDR-corrected map
group_fdr = dataset.load_group_level_contrast(
    contrast='faceVsBaseline',
    correction='fdr'
)

# Display both
fig, axes = plt.subplots(1, 2, figsize=(15, 5))

plotting.plot_stat_map(
    group_contrast,
    threshold=3.1,
    title='Uncorrected (z > 3.1)',
    axes=axes[0],
    cut_coords=(0, 0, 0)
)

plotting.plot_stat_map(
    group_fdr,
    threshold=0.05,
    title='FDR Corrected (q < 0.05)',
    axes=axes[1],
    cut_coords=(0, 0, 0)
)

plt.tight_layout()
plt.show()

Statistical Metadata

Analysis Parameters

Each GLM analysis includes metadata documenting the analysis parameters:

{
    "AnalysisLevel": "first_level",
    "Model": {
        "Type": "GLM",
        "HRF": "spm",
        "DriftModel": "cosine",
        "HighPassFilter": 0.01,
        "NoiseModel": "ar1"
    },
    "Contrasts": {
        "faceVsBaseline": {
            "Description": "Face stimuli versus baseline",
            "Weights": [1, 0, 0, -1]
        },
        "faceVsScene": {
            "Description": "Face stimuli versus scene stimuli",
            "Weights": [1, -1, 0, 0]
        }
    },
    "Software": {
        "Name": "Nilearn",
        "Version": "0.10.0",
        "Container": "nilearn/nilearn:0.10.0"
    },
    "Confounds": [
        "trans_x", "trans_y", "trans_z",
        "rot_x", "rot_y", "rot_z",
        "global_signal", "csf", "white_matter"
    ]
}