SNM_1000_V1.0

Release v1.0 — Pretrained SNM1000 Model

This is the first public release of the Spectral Normative Model (SNM1000) for cortical thickness, accompanying the manuscript:

Mansour L., S., et al. (2025). Spectral Normative Modeling of Brain Structure. medRxiv. DOI: 10.1101/2025.01.16.25320639

---

📦 Release Assets

• pretrained_SNM_1000_V1.0.tar.gz — Pretrained SNM1000 model for cortical thickness, fitted on >78,000 healthy brains from 30 large-scale neuroimaging datasets.

---

🧠 Model Description

The SNM1000 is a Spectral Normative Model trained on cortical thickness data spanning the human lifespan. It encodes cortical phenotypes using 1000 connectome eigenmodes and models normative trajectories as a function of age, sex, and site using Hierarchical Bayesian Regression.

The pretrained model can be used to:

• Generate normative cortical thickness trajectories for any parcellation
• Compute individual deviation scores from the healthy normative range
• Apply normative adaptation (fine-tuning) to new sites or clinical samples

---

🚀 Usage

Install the SpectraNorm package:

pip install spectranorm --upgrade

Decompress the model files:

tar -xzvf pretrained_SNM_1000_V1.0.tar.gz

Load the model in Python:

from spectranorm import snm

# load decompressed model (replace with actual path)
snm_1000 = snm.SpectralNormativeModel.load_model(
    "/<path-to-extracted-directory>/pretrained_SNM_1000_V1.0/"
)

For users who want to better understand the underlying methodology, refer to the SpectraNorm tutorials for step-by-step walkthroughs (more tutorials coming soon).

The following notebooks in the repository demonstrate example usage of the pretrained model:

• Notebook 7 — Cortical Growth Gradients: Map cortical growth gradients across the lifespan using the pretrained SNM
• Notebook 8 — Clinical (AD) Evaluations: Apply the pretrained SNM to a new clinical sample, including site adaptation and individual deviation mapping
• Notebook 9 — Data Sharing: Generate normative charts across alternative cortical parcellations using the pretrained SNM

---

📄 License

This release is subject to the dual licensing terms described in the LICENSE file.