Skip to main navigation Skip to search Skip to main content

Early Prediction of Tongue Image Diabetes Based on Image Segmentation and Liquid Neural Network

    Activity: Supervision › Completed SURF Project

    Description

    The scanning of diabetes strains healthcare systems. 1ongue images facilitate non-invasive auxiliary screening, yet raw photographs are compromised by background interference. We deployed zero-shot SAM2 segmentation for ResNet18classification and benchmarked diverse prompting paradigms. The optimal SAM2 pipeline elevated sensitivity from 0.892 to 0.973 while preserving specificity, enabling scalable tongue-image-assisted diabetes screening for community deployment
    Period29 Jun 2026 → 29 Aug 2026

    Keywords

    • Zero-shot Segmentation
    • SAM2
    • Diabetes Screening
    • Non-invasive Screening
    • Tongue Image