Selected work

Representative Papers

Towards clinical adoption of voice and speech as measures of health: the need for harmonization

Sep 23, 2026

This study addresses the limited reproducibility of vocal biomarkers caused by heterogeneous data and processing pipelines, which severely impedes their clinical translation. By systematically reviewing the full lifecycle of voice-based health measurement, this work proposes a standardization pathway anchored in unified acoustic metric definitions. Specifically, it establishes a minimal core metric set encompassing respiratory and phonatory dimensions with clearly defined physiological mappings, while integrating acoustic feature computation, machine learning modeling, and clinical interpretation into a cohesive framework. The research provides reproducible metric definitions alongside practical implementation schemes, thereby laying a foundational framework for the standardization and clinical deployment of digital vocal biomarkers.

0 citationsRead paper

Simultaneous Image Quality Improvement and Artefacts Correction in Accelerated MRI

Nov 28, 2025

In accelerated MRI with undersampling, diagnostic reliability is compromised by degraded image quality, concurrent noise, and motion artifacts. To address this, we propose USArt—a novel deep learning framework that jointly models undersampled reconstruction and multi-class artifact correction (noise and motion). USArt employs a dual-branch collaborative architecture: one branch optimizes image fidelity, while the other suppresses artifacts. Specifically designed for Cartesian 2D brain MRI, it supports diverse undersampling patterns. Experiments demonstrate that, at up to 5× acceleration, USArt significantly improves signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR), effectively eliminates artifacts, and maintains strong robustness across multiple degradation scenarios—including varying noise levels, motion magnitudes, and sampling patterns. To our knowledge, USArt is the first method to unify undersampling reconstruction and heterogeneous artifact correction in a single end-to-end trainable model. It establishes a new paradigm for rapid, high-fidelity clinical MRI acquisition.

0 citationsRead paper

Recent publications

Latest Papers

Towards clinical adoption of voice and speech as measures of health: the need for harmonization

Sep 23, 2026

This study addresses the limited reproducibility of vocal biomarkers caused by heterogeneous data and processing pipelines, which severely impedes their clinical translation. By systematically reviewing the full lifecycle of voice-based health measurement, this work proposes a standardization pathway anchored in unified acoustic metric definitions. Specifically, it establishes a minimal core metric set encompassing respiratory and phonatory dimensions with clearly defined physiological mappings, while integrating acoustic feature computation, machine learning modeling, and clinical interpretation into a cohesive framework. The research provides reproducible metric definitions alongside practical implementation schemes, thereby laying a foundational framework for the standardization and clinical deployment of digital vocal biomarkers.

0 citationsRead paper

Simultaneous Image Quality Improvement and Artefacts Correction in Accelerated MRI

Nov 28, 2025

In accelerated MRI with undersampling, diagnostic reliability is compromised by degraded image quality, concurrent noise, and motion artifacts. To address this, we propose USArt—a novel deep learning framework that jointly models undersampled reconstruction and multi-class artifact correction (noise and motion). USArt employs a dual-branch collaborative architecture: one branch optimizes image fidelity, while the other suppresses artifacts. Specifically designed for Cartesian 2D brain MRI, it supports diverse undersampling patterns. Experiments demonstrate that, at up to 5× acceleration, USArt significantly improves signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR), effectively eliminates artifacts, and maintains strong robustness across multiple degradation scenarios—including varying noise levels, motion magnitudes, and sampling patterns. To our knowledge, USArt is the first method to unify undersampling reconstruction and heterogeneous artifact correction in a single end-to-end trainable model. It establishes a new paradigm for rapid, high-fidelity clinical MRI acquisition.

0 citationsRead paper