Physics-guided synthetic ultrasound aids skin layer segmentation
Physics-Guided Synthetic High-Frequency Ultrasound Generation for Skin Layer Segmentation
Machine LearningComputer Vision and Pattern Recognition
Summary
High-frequency ultrasound helps see the layers of skin without cutting it, but there isn't enough detailed data showing all the skin layers. The authors created fake ultrasound images using physics-based models of skin layers to generate lots of training data. When a computer learned from these fake images and then from real ones, it got better at identifying skin layers in real ultrasound images. This approach shows promise but needs more work to make the fake images look more like real ones.
What this means in practice
- •For medical imaging developers: Generate synthetic skin ultrasound images with dense labels to improve training of skin layer segmentation models.
- •For dermatology device makers: Enhance automated analysis of skin ultrasound scans by pretraining models on physics-based synthetic data before using real patient data.$Commercial implications: Enables more accurate skin diagnostic tools by supplementing limited annotated ultrasound data, improving commercial skin imaging devices.
Authors
Junkyung ju, Kyungho Yoon, Minwoo Shin
Abstract
High-frequency ultrasound (HFUS) enables noninvasive visualization of superficial skin structures, but automated skin-layer analysis is limited by the scarcity of densely annotated data. Existing real HFUS datasets commonly provide annotations for superficial targets such as the epidermis and subepidermal low-echogenic band (SLEB), while dense labels for deeper structures such as dermis, subcutaneous tissue, fascia, and muscle are rarely available. We propose a physics-guided synthetic HFUS generation framework for skin layer segmentation. The framework constructs multilayer acoustic skin phantoms, assigns layer dependent acoustic properties, and uses k-Wave simulation to generate paired synthetic HFUS images, dense layer masks, and simulation metadata. To evaluate whether the generated data provide transferable supervision, we use it for downstream segmentation pretraining and fine-tune the models on real Mendeley HFUS data. Synthetic pretraining followed by real fine-tuning achieved real-domain performance comparable to real-only training and improved mean Dice/IoU in three of four evaluated trainable architectures. These results suggest that physics-guided synthetic HFUS images contain transferable anatomical and textural cues for real-domain skin layer segmentation, although further reduction of the synthetic-real appearance gap is needed to enable greater gains. The code and data are available at: https://github.com/Finn-02/synthetic-hfus-skin-layer-segmentation.