arXiv subjects

Gabriela Hossu

Publications and source records attributed to Gabriela Hossu.

2 recordsLinked to original sources

arXiv2026

Articulatory movements influence electromagnetic wave transmission through the vocal tract

This study experimentally validates a numerical model of electromagnetic propagation through the human head during the pronunciation of different vowels, with the goal of improving our understanding of the underlying physical phenomena. A realistic finite element model was created from magnetic resonance images acquired while pronouncing the vowels /a/, /i/, and /u/. The model was validated against scattering matrix measurements obtained from two subjects whose geometries were modeled. Despite several potential sources of discrepancy, the simulations and measurements showed good qualitative agreement, confirming the validity of the approach. Similar transmission coefficient patterns were observed across subjects for the same vowels. Within the investigated frequency range of (1-6~GHz), the electric field exhibited a Mie scattering pattern. Local minima and maxima in the transmission coefficient, characterizing different articulatory configurations, were correlated with local variations in the electric field amplitude. The transmission coefficient's shape results from an interplay between resonance patterns and antenna placement, while the degree of mouth opening influences the shape of scattering modes. Although technically challenging, this numerical approach proved effective for studying electromagnetic propagation in the human head. The resulting robust numerical model and improved understanding of the underlying physics are expected to facilitate the development of radio-frequency-based silent speech interfaces.

physics.app-ph↗

arXiv2026

Dataset-Aware Cold-Start Active Learning for Annotation-Efficient 3D Medical Image Segmentation

Deep learning for 3D medical image segmentation requires extensive manual annotations, a major bottleneck in volumetric medical imaging. Active learning aims to reduce this burden by selecting informative samples for annotation, but most methods assume that an initial labeled set is already available. This leaves the cold-start problem largely unresolved: how to select the first volumes from a fully unlabeled pool before any task-specific model is trained. We propose CSCS, a Curriculum-Stratified Cold-Start framework that adapts initial sample selection to the structure of the unlabeled dataset. CSCS combines two self-supervised, label-free signals: local typicality, measuring representativeness in the embedding space, and reconstruction-based uncertainty, used as a proxy for sample difficulty. These signals are combined through a weighted geometric score, where the weighting is determined by a closed-form pacing rule based on the effective annotation budget and the Difficulty-Coverage Ratio, a pool-level statistic measuring the alignment between difficulty and representativeness. We evaluate CSCS on four 3D medical image segmentation benchmarks: BraTS, FeTA, Spleen, and an in-house fetal MRI dataset. Using nnU-Net as downstream segmentation model, CSCS shows consistently competitive performance across datasets and annotation budgets, with the strongest gains in low-to-mid annotation regimes. These results suggest that dataset-aware cold-start initialization can improve the robustness of active learning for 3D medical image segmentation by adapting sample selection to the geometry of the unlabeled pool.

eess.IV↗