Welcome to Psychiatryai.com: Latest Evidence - RAISR4D

Estimated reading time for CME/CPD: 2 mins

FroHyNet: frozen temporal priors and trainable hypergraph fusion for brain time-series classification

AI Summary
  • FroHyNet combines a frozen MantisV2 temporal encoder with a trainable spatio-temporal hypergraph branch, using BM-TopK to selectively residual inject spatial corrections.
  • Outperforms competing methods on five fNIRS and fMRI cohorts, achieving highest average accuracy and AUC while requiring far fewer trainable parameters.
  • Ablations show dataset-dependent complementarity between streams and that BM-TopK fusion outperforms dense and recurrent fusion on the HC-D dataset.
Summarise with AI (MRCPsych/FRANZCP)

Front Med (Lausanne). 2026 Oct 7;13:1966334. doi: 10.3389/fmed.2026.1966334. eCollection 2026.

ABSTRACT

BACKGROUND: Functional brain time-series classification is constrained by small clinical cohorts. Handcrafted features are sample-efficient but discard temporal detail; end-to-end networks must estimate many temporal parameters from limited labels; and frozen time-series foundation models reduce that estimation burden but typically encode each channel independently, leaving coordinated activity across brain regions unmodeled.

METHODS: We introduce the Frozen Temporal-Hypergraph Network (FroHyNet), an asymmetric parallel framework that preserves channel-wise temporal representations from a frozen MantisV2 encoder while a trainable spatio-temporal hypergraph branch learns topology-aware spatial corrections. Bottleneck-mix Top-K (BM-TopK) then residual-injects these corrections only along selected channel and latent-feature axes, treating the spatial stream as a selective update rather than a dense overwrite of the temporal representation. We conduct five-fold cross-validation across two functional near-infrared spectroscopy (fNIRS) and three functional magnetic resonance imaging (fMRI) datasets.

RESULTS: FroHyNet achieves the best average accuracy and area under the receiver-operating-characteristic curve (AUC) among methods evaluated on all five datasets, while keeping the trainable-parameter count well below those of from-scratch temporal networks. Gains are concentrated on fNIRS, with more modest improvements on the smaller fMRI cohorts. Branch ablations indicate dataset-dependent complementarity between the two streams, and fusion ablations on HC-D favor BM-TopK over the evaluated dense and recurrent alternatives.

PMID:42908832 | PMC:PMC13683208 | DOI:10.3389/fmed.2026.1966334

Document this CPD

Share Evidence Blueprint

QR Code

Save to Google Notes

Search Google Scholar

⭐ My Revision List

close chatgpt icon
ChatGPT

Enter your request.

← →
RAISR4D CME/CPD Evidence Nodes
Swipe to navigate RAISR4D CME/CPD evidence nodes.
CME/CPD