Dual Co-Train: Cross-Dataset Ultrasound Tongue Segmentation Under Extreme Data Scarcity
Researchers developed a lightweight AI model that segments ultrasound tongue contours with high accuracy using just five labeled images, adapting to new datasets without additional annotations.
- Uses only five labeled images to pretrain a lightweight UltraUNet model for ultrasound tongue segmentation.
- Adapts to new unlabeled datasets via iterative pseudo-label refinement and unreliable mask filtering.
- Eliminates the need for target domain annotations, addressing data scarcity in clinical settings.
- Demonstrates robustness to probe variability and acquisition noise across cross-dataset scenarios.
A new study introduces Dual Co-Train, a source-free domain adaptation framework designed to improve ultrasound tongue segmentation under extreme data scarcity. The approach leverages a lightweight UltraUNet backbone, starting from a checkpoint pretrained on only five labeled source images. This simulates a constrained source model, which is then adapted to a fully unlabeled target domain through iterative pseudo-label refinement and unreliable mask filtering. The method addresses challenges like probe variability and acquisition noise, which typically degrade model generalization across datasets.
The framework eliminates the need for target domain annotations, making it highly practical for clinical and research settings where labeled data is scarce. By iteratively refining pseudo-labels, the model achieves robust segmentation performance even when trained on minimal labeled examples. This could significantly reduce the time and cost associated with collecting and annotating ultrasound tongue imaging data.
The research highlights the potential of lightweight AI models in medical imaging, particularly in scenarios where computational resources and labeled data are limited. The proposed method could pave the way for more accessible and scalable solutions in speech therapy, linguistics, and related fields.
Provides a practical framework for medical imaging AI with minimal labeled data requirements.
Reduces costs and time for developing ultrasound-based speech and language technologies.
Offers a case study in domain adaptation and lightweight model design for medical imaging.
Advances AI-driven medical imaging with potential applications in speech therapy and linguistics.
- Domain adaptation
- A machine learning technique that adapts a model trained on one dataset to perform well on a different but related dataset.
- Pseudo-labeling
- A semi-supervised learning method where the model generates its own labels for unlabeled data and uses them for training.
- Ultrasound tongue segmentation
- The process of identifying and outlining the tongue's contours in ultrasound images for speech and language analysis.
Artificial intelligence acts as an 'ideological chameleon' and may deepen political polarization - Phys.org
AI ResearchHow Much Memory Does Your Agent Actually Need?
From Corpora to Co-Evolving Capabilities: Capability-Centric Data Design for Generalist Image Generation
On the Fragility of Self-Improving Agents: Variance, Task Order, and Underspecification
Delegation Asymmetry in Agentic Recommender Systems: Measuring Two-Sided Receptivity in Online Dating
ProgrammingMy QUIC transport had never once been executed. Here's what happened when I ran it.
A developer discovered three critical bugs and flawed semantics in a QUIC-based protocol after finally executing it, despite never running it before.
Open Sourceopen-doc: Letting Antigravity and Other Coding Agents Fully Own Document Layout and Generation
A new open-source tool called open-doc enables AI coding agents to autonomously handle document layout and generation tasks.
Expanded curriculum includes AI~focused learning - James Madison University
James Madison University is expanding its curriculum to include AI-focused learning modules for students across disciplines.
Artificial intelligence boosts automated biolabs - Knowable Magazine
AI is enhancing automated biolabs by improving efficiency and accuracy in experiments. Knowable Magazine reports on these advancements.
Broadcom's Artificial Intelligence (AI) Revenues Are Forecast to Exceed $100 Billion in 2027: Should You Buy the Dip? - The Motley Fool
Analysts project Broadcom's artificial intelligence revenue could surpass $100 billion by 2027, driven by demand for its AI infrastructure solutions. The forecast suggests significant growth for the semiconductor giant in the AI sector.
Broadcom's Artificial Intelligence (AI) Revenues Are Forecast to Exceed $100 Billion in 2027: Should You Buy the Dip? - Yahoo Finance
Broadcom’s AI-related revenue is projected to surpass $100 billion by 2027, driven by demand for AI accelerators and custom chips.