Enhancing Few-Shot Medical Image Segmentation With Med-CAP: A Step Toward Personalized Healthcare in the Digital Twin Era

The advancement of Healthcare 5.0 and medical Digital Twin systems increasingly relies on accurate anatomical modeling, real-time physiological monitoring, and personalized treatment simulation. These capabilities rely on effective medical image segmentation. However, the lack of annotated medical data and the diversity of clinical imaging severely restrict the applicability of conventional fully supervised models. To address these practical limitations, we propose Med-CAP, a Medical Cross-image Adaptive Prior Segmentation Network designed for real-world low-annotation settings. Med-CAP integrates self-attention–based context encoding, cosine-similarity prior generation, a lightweight prior-guided fusion decoder, and cross-image adaptive multi-head attention to improve structural correspondence between support and query images with minimal supervision. By using just a single annotated example for reliable segmentation, Med-CAP offers a scalable and privacy-friendly solution. It is well-suited for deployment in smart hospitals, remote healthcare systems, and medical Digital Twin platforms. Experiments on abdominal CT and MRI datasets demonstrate that Med-CAP not only achieves superior accuracy compared with state-of-the-art few-shot methods but also exhibits strong robustness across modalities, highlighting its value for next-generation intelligent healthcare applications.