HypTI
HypTI is a hyperbolic-geometry-based multimodal fusion framework for dementia detection. It uniformly processes clinical tabular data, MRI images, and cross-modal interactions in Lorentz hyperbolic space, leveraging the hierarchical representation capacity of hyperbolic space to model the hierarchical relationships among clinical variables and the nested patterns of brain region structures.
Timeline (last 90 days)
The paper proposes the HypTI framework, which introduces hyperbolic geometry into joint tabular-imaging modeling for dementia detection.
HypTI handles three kinds of information in a unified way in Lorentz hyperbolic space: clinical tabular data, MRI imaging and cross-modal interactions.
HypTI learns a Lorentz prototype representation for each tabular assessment item, mapping clinical scores to points in hyperbolic space, so that hierarchical relationships are naturally reflected through prototype positions.
HypTI designs an adaptive fusion module that generates subject-specific scale and shift coefficients to dynamically modulate and integrate multimodal embeddings, with a mechanism similar to the FiLM idea.
The paper conducted experiments on the OASIS-3 dataset, and HypTI achieved strong performance compared with representative methods.
The paper abstract does not disclose specific metric values or ablation details, and the magnitude of the gain brought by hyperbolic space still needs to be supported by the full paper's data.
The initialization strategy of the Lorentz prototypes, the degree of standardization of ROI definitions, and the model's generalization ability on more heterogeneous cohorts are issues worth attention in follow-up work.
All Facts (7)
The paper proposes the HypTI framework, which introduces hyperbolic geometry into joint tabular-imaging modeling for dementia detection.
50%UnverifiedHypTI handles three kinds of information in a unified way in Lorentz hyperbolic space: clinical tabular data, MRI imaging and cross-modal interactions.
50%UnverifiedHypTI learns a Lorentz prototype representation for each tabular assessment item, mapping clinical scores to points in hyperbolic space, so that hierarchical relationships are naturally reflected through prototype positions.
50%UnverifiedHypTI designs an adaptive fusion module that generates subject-specific scale and shift coefficients to dynamically modulate and integrate multimodal embeddings, with a mechanism similar to the FiLM idea.
50%UnverifiedThe paper conducted experiments on the OASIS-3 dataset, and HypTI achieved strong performance compared with representative methods.
50%UnverifiedThe paper abstract does not disclose specific metric values or ablation details, and the magnitude of the gain brought by hyperbolic space still needs to be supported by the full paper's data.
50%UnverifiedThe initialization strategy of the Lorentz prototypes, the degree of standardization of ROI definitions, and the model's generalization ability on more heterogeneous cohorts are issues worth attention in follow-up work.
50%