The problem
Image-based analysis of carotid intima-media thickness (IMT) requires selecting suitable images and identifying the vessel-wall region. Mobile use also requires evaluation of computational and memory demands alongside image-analysis performance.
The research idea
MobileNetV3 first classifies images relevant to IMT analysis, and SegNet then segments the IMT region. The abstract also discusses 8-bit quantization, pruning and knowledge distillation for model compression.
What was evaluated
The official BIOIMAGING 2025 abstract reports evaluation on 190 annotated carotid ultrasound images. The reported technical evaluation concerns image classification, segmentation and execution demands.
Considering applications
Separating image selection from segmentation offers a design reference for portable ultrasound analysis. A device-specific evaluation can compare accuracy changes and execution demands after compression under the target acquisition conditions.
Original research
Efficient Automatic IMT Region Extraction Using MobileNetV3 and SegNet for Mobile Carotid Ultrasound Imaging
Usami, Hiroyasu, Nakayama, Natsuki, Arakawa, Naoko
BIOIMAGING 2025, 2025 · BIOIMAGING25-367: Abstract, 20 min oral presentation