E4-a1 - RG-UNet: Automated Muscle Ultrasound Segmentation Using the U-Net Architecture Augmented with Monogenic Phase Asymmetry Maps
- Event
- 2025 ICU PADERBORN - 9th International Congress on Ultrasonics
2025-09-21 - 2025-09-25
Paderborn - Band
- Lectures
- Chapter
- E4-a - Machine Learning and Statistical Methods in Acoustics
- Author(s)
- P. Engl, J. Schwerdt, K. -V. Jenderka - Merseburg University of Applied Sciences, Merseburg (Germany), H. S. Aghamiry - Charité-Universitätsmedizin Berlin, Berlin (Germany)
- Pages
- 291 - 294
- DOI
- 10.5162/Ultrasonic2025/E4-a1
- ISBN
- 978-3-910600-08-9
- Price
- free
Abstract
This study presents a novel deep learning approach for automated image segmentation of dynamic ultrasound images of the vastus lateralis muscle, which addresses key challenges like data availability and image variability. The proposed RG-UNet augments a standard lightweight U-Net architecture with a secondary input channel derived from monogenic phase asymmetry analysis, providing the network with intensity-invariant structural information for tissue boundaries and muscle fascicles and by this achieving superior segmentation performance.