TReND: Transformer derived features and Regularized NMF for neonatal functional network Delineation

TL;DR

Neonatal functional network delineation is hindered by rapid maturation and the lack of a standardized atlas, making adult networks unsuitable. The authors introduce TReND, a self-supervised transformer-based autoencoder that learns temporal features and refines them with regularized nonnegative matrix factorization (RNMF) followed by KMeans clustering to produce robust neonatal FN parcellations. The framework integrates confidence-adaptive attention and geodesic spatial encodings to boost temporal feature extraction and spatial coherence, and is validated on simulated data, the dHCP neonatal cohort, and cross-dataset HCP-YA with a Yeo atlas reference, achieving high reproducibility and improved spatial contiguity and functional homogeneity. The resulting 7- and 19-network neonatal atlases offer a standardized resource for studying perinatal brain development and disorders.

Abstract

Precise parcellation of functional networks (FNs) of early developing human brain is the fundamental basis for identifying biomarker of developmental disorders and understanding functional development. Resting-state fMRI (rs-fMRI) enables in vivo exploration of functional changes, but adult FN parcellations cannot be directly applied to the neonates due to incomplete network maturation. No standardized neonatal functional atlas is currently available. To solve this fundamental issue, we propose TReND, a novel and fully automated self-supervised transformer-autoencoder framework that integrates regularized nonnegative matrix factorization (RNMF) to unveil the FNs in neonates. TReND effectively disentangles spatiotemporal features in voxel-wise rs-fMRI data. The framework integrates confidence-adaptive masks into transformer self-attention layers to mitigate noise influence. A self supervised decoder acts as a regulator to refine the encoder's latent embeddings, which serve as reliable temporal features. For spatial coherence, we incorporate brain surface-based geodesic distances as spatial encodings along with functional connectivity from temporal features. The TReND clustering approach processes these features under sparsity and smoothness constraints, producing robust and biologically plausible parcellations. We extensively validated our TReND framework on three different rs-fMRI datasets: simulated, dHCP and HCP-YA against comparable traditional feature extraction and clustering techniques. Our results demonstrated the superiority of the TReND framework in the delineation of neonate FNs with significantly better spatial contiguity and functional homogeneity. Collectively, we established TReND, a novel and robust framework, for neonatal FN delineation. TReND-derived neonatal FNs could serve as a neonatal functional atlas for perinatal populations in health and disease.

Figures (5)

• Figure 1: Schematic representation of end-to-end functional parcellation framework. A. Processed Blood Oxygenation Level Dependent (BOLD) fMRI data. B. Feature extraction using transformer-based autoencoder to convert BOLD signals into low-dimensional embeddings. C. Correlation matrix from the transformer-based feature embeddings. D. Clustering using RNMF $+$ KMeans are performed to generate parcellation of different regions. E. Parcellation shows different regions based on correlation matrix.
• Figure 2: Performance evaluation of TReND versus traditional parcellation methods on simulated data. A.Left panel: The cartoon illustrates how TReND extracts features and preserves more fine-grained details. Right panel: Comparison of feature extraction techniques: Principal component analysis (PCA), Uniform manifold approximation and projection (UMAP), Tensor decomposition (TD), and TReND. B.Left panel: The cartoon shows how TReND produces spatially coherent clusters. Right panel: Comparison of clustering methods: kernel-PCA (kPCA), Independent component analysis (ICA), NMF, and TReND.
• Figure 3: Coarse parcellation of 7 functional networks in neonates derived from a 300-subject dHCP cohort. A. Stability analysis of the clustering algorithm identifies 7 and 19 networks as robust estimates. B. Neonatal cortical parcellation of 7 FNs. C. Confidence map representing the reliability of the 7 neonatal FNs parcellation. D. Discovery and replication of the 7 neonatal FNs cortical parcellation. E. Dice score evaluation FN-wise. F. Comparison of primary and higher-order FNs in TReND-derived neonate and adult atlas ref3. (Abbreviation: C.: Component)
• Figure 4: Fine-grained parcellation of 19 functional networks in neonates derived from a 300-subject dHCP cohort. A. Neonatal cortical parcellation of 19 FNs. B. Confidence map representing the reliability of the 19 neonatal FNs parcellation. C. Discovery and replication of the 19 neonatal FNs cortical parcellation. D. Dice score evaluation of FN-wise via bootstrap analysis, assessing the consistency of the parcellation.
• Figure 5: Validation of TReND's performance on the HCP-YA dataset. Consistency of the 7 FNs identified by TReND compared to the Yeo 7-network atlas evaluated with dice.