Back to all news
Deep Learning Model Enables 3D Handheld Photoacoustic-Ultrasound Imaging Without Sensors
Pusan National University develops MoGLo-Net, an AI model that reconstructs 3D images from handheld 2D photoacoustic and ultrasound scans without external sensors.
Key Details
- 1MoGLo-Net uses deep learning to track handheld ultrasound transducer motion from tissue speckle data, eliminating need for external tracking hardware.
- 2Combines ResNet-based encoder and LSTM-based motion estimator for accurate motion tracking and 3D reconstruction.
- 3Validated using both proprietary and public datasets, outperforming state-of-the-art methods on all metrics.
- 4Successfully achieved 3D blood vessel reconstructions from combined ultrasound and photoacoustic data.
- 5Published June 13, 2025, in IEEE Transactions on Medical Imaging (DOI: 10.1109/TMI.2025.3579454).
- 6Innovation aims to make advanced 3D imaging safer, more accurate, and accessible without costly hardware.
Why It Matters
MoGLo-Net's ability to reconstruct 3D volumes from standard handheld imaging without external sensors could democratize high-end ultrasound and photoacoustic imaging, especially important in resource-limited settings. This advance may enable more effective diagnosis and real-time imaging guidance, expanding access to advanced imaging technologies while reducing costs and complexity of equipment.
Source
EurekAlert
Related News
•EurekAlert
AI Model Improves Prediction of Knee Osteoarthritis Progression Using MRI and Biomarkers
A new AI-assisted model that combines MRI, biochemical, and clinical data improves predictions of worsening knee osteoarthritis.
•EurekAlert
AI Model Predicts Multiple Genetic Markers from Colorectal Pathology Slides
Researchers developed and validated an AI model that simultaneously detects multiple genetic markers in colorectal cancer tissue slides.
•EurekAlert
Photonic Chip Enables Versatile Neural Networks for Imaging and Speech AI
Chinese scientists have developed a reconfigurable integrated photonic chip capable of running diverse neural networks, including those for image and speech processing, with high efficiency.