Skip to content

Input Data Format

PRISME accepts neuroimaging data as MATLAB .mat files with a specific structure. This section describes the required format.

Overview

Your dataset file must contain three components:

  • brain_data - The neuroimaging data (FC matrices or voxel activations)
  • study_info - Metadata about the dataset
  • outcome - Study definitions and test specifications

brain_data

The brain_data component is a MATLAB struct where each field represents a condition or task (e.g., REST, EMOTION, SOCIAL).

Each condition field contains:

  • data: Flattened neuroimaging data where each column represents one subject
  • For FC data: Each row is an edge measuring the correlation between two ROI time series
  • For voxel data: Each row is a voxel measuring activation (referred to as nodes in graph representation)
  • sub_ids: Subject IDs corresponding to each column in the data matrix
  • mask: (Optional - can be stored in study_info if consistent across tasks) A logical structure that maps flattened variables back to their original spatial positions (ROI pairs for connectivity matrices, brain coordinates for voxels)

study_info Structure

The study_info component contains metadata describing your dataset.

Required fields:

  • dataset - Dataset identifier (e.g., 'HCP', 'ABCD')
  • map - Data type (e.g., 'functional_connectivity', 'activation')
  • mask - A logical structure that maps flattened variables to their original spatial positions

outcome Structure

The outcome component defines the studies PRISME will analyze. Each study is stored as a struct, and the number of elements in outcome equals the number of studies.

Required fields for all test types:

  • sub_ids - Subject identifiers included in this study (can also be stored in brain_data)
  • reference_condition - Baseline condition (e.g., 'REST')
  • category - Study category label

Additional fields by test type:

For correlation analyses: - score - Continuous measures (e.g., age, behavioral scores) - score_label - Description of the measure

For condition contrasts: - contrast - For studies analyzing the difference between two conditions, this specifies both conditions to contrast

Test Type Inference

PRISME automatically infers the test type (one-sample, two-sample, or correlation) from the outcome structure. The function infer_test_from_data uses the following logic:

  1. One-sample t-test: If score contains only one unique value, or if contrast specifies a single condition
  2. Correlation test: If score is numeric with more than 2 unique values
  3. Two-sample or paired t-test: If score has exactly 2 unique values, or if contrast specifies 2 conditions:
  4. Calculates subject overlap between the two groups
  5. If overlap ≥ unique subjects → One-sample t-test (paired design)
  6. If overlap < unique subjects → Two-sample t-test (independent groups)

Converting Your Data

If your data is not in PRISME format, you'll need to write a conversion script that matches the description above. Please follow the data description. We also provide dummy datasets that are used for testing in the repository. Please check test named datasets under /data/ as a reference to the input data format.

Example Datasets

Preprocessed HCP and ABCD data in PRISME format can be obtained from the authors upon verification of approved data access from the respective repositories.