Sign in Subscribe
Artificial intelligence

AI-Assisted Anatomical Structure Recognition and Segmentation Improves Diagnostic Sonography

Strategic Insights

— 1 min read

Researchers have developed an AI-based framework, MaskHybrid, that enhances the accuracy and efficiency of anatomical structure recognition and segmentation in abdominal ultrasonography. The framework uses a hybrid architecture combining mamba-transformer and deep neural networks (DNNs) to capture long-range spatial dependencies and contextual information. The study used a private dataset of 34,711 abdominal ultrasound images and achieved a mean average precision (mAP) score of 74.13% for anatomical landmarks segmentation.

Key Takeaways:

  • The study developed an AI-based framework, MaskHybrid, that enhances the accuracy and efficiency of anatomical structure recognition and segmentation in abdominal ultrasonography.
  • The framework uses a hybrid architecture combining mamba-transformer and deep neural networks (DNNs) to capture long-range spatial dependencies and contextual information.
  • The study used a private dataset of 34,711 abdominal ultrasound images and achieved a mean average precision (mAP) score of 74.13% for anatomical landmarks segmentation.
  • MaskHybrid outperformed baselines across most organ and lesion types and effectively segmented challenging anatomical structures.
  • The framework exhibited a significantly shorter inference time (0.120 ? 0.013 s), achieving 2.5 times faster than large-sized AI models of similar size.
  • The study concluded that MaskHybrid can facilitate improved medical image interpretation and near real-time diagnostic sonography that meets clinical needs.
  • The framework can be applied to various medical imaging tasks, including computer-aided diagnosis, image denoising, and image segmentation.

Statistics:

  • 74.13%: Mean average precision (mAP) score achieved by MaskHybrid for anatomical landmarks segmentation in abdominal ultrasound images.
  • 2.5 times: Factor by which MaskHybrid was faster than large-sized AI models of similar size.
  • 0.120 ? 0.013 s: Inference time achieved by MaskHybrid.
  • 34,711: Number of abdominal ultrasound images used in the study.
  • 2,063: Number of patients in the study.

Sources:

  • NewsRx. Data on Artificial Intelligence Discussed by Researchers at Industrial Technology Research Institute (AI-assisted anatomical structure recognition and segmentation via mamba-transformer architecture in abdominal ultrasound images). Robotics & Machine Learning. August 25, 2025; p 70.

Read more