Equivariant Spherical CNNs for Neonatal Diffusion MRI

This repository provides the official implementation of our paper:

Equivariant Spherical CNNs for Accurate Fiber Orientation Distribution Estimation in Neonatal Diffusion MRI with Reduced Acquisition Time
Haykel Snoussi and Davood Karimi
Department of Radiology, Boston Children’s Hospital & Harvard Medical School

Summary

We introduce a geometric deep learning framework based on rotationally equivariant Spherical CNNs (sCNNs) to estimate Fiber Orientation Distributions (FODs) from neonatal diffusion MRI (dMRI), using only 30% of the full diffusion acquisition protocol. This approach enables faster and more practical neonatal imaging. The model was trained and evaluated on 43 neonatal dMRI datasets from the Developing Human Connectome Project (dHCP).

Highlights

• Uses only 30% of the dHCP acquisition protocol — reducing scan time and motion artifacts.
• Employs SO(3)-equivariant spherical convolutions to preserve rotational symmetries of diffusion signals.
• Incorporates a shell-attention mechanism for adaptive fusion across b-value shells.
• Optimized with a spatial-domain loss function to prioritize perceptually meaningful FOD reconstructions.
• Delivers superior tractography compared to both standard MLP and MSMT-CSD methods.

Method Overview

Figure: Overview of the full data processing pipeline and sCNN architecture.

---

FOD Estimation Accuracy

Figure: Comparison of FODs predicted by MLP, sCNN (30% data), and MSMT-CSD ground truth.

---

Tractography Analysis

Figure: Tractography results using FODs from MLP, sCNN, and MSMT-CSD.

---

📄 Citation

If you use this work, please cite:

@article{snoussi2025equivariant,
  title={Equivariant Spherical CNNs for Accurate Fiber Orientation Distribution Estimation in Neonatal Diffusion MRI with Reduced Acquisition Time},
  author={Snoussi, Haykel and Karimi, Davood},
  journal={arXiv preprint arXiv:2504.01925},
  year={2025}
}