From Image Synthesis to Digital Twins: A Paradigm Shift in AI-Driven Medical Image Computing

Abstract

The integration of generative artificial intelligence (AI) in medical imaging represents a transformative paradigm shift from traditional discriminative models to sophisticated data synthesis frameworks, ultimately advancing toward the vision of patient-specific Digital Twins. This comprehensive review examines the technical evolution, clinical applications, and future perspectives of generative AI in medical image generation. The review categorizes clinical applications into three domains: (1) diagnostic enhancement through image quality improvement and cross-modal synthesis, (2) pathology simulation and prognostic prediction via longitudinal disease modeling, and (3) medical research and education through privacy-preserving data augmentation. We identify critical challenges impeding clinical translation, including limited evaluation metrics that fail to capture clinical validity, fairness and robustness issues arising from biased training data, lack of clinical interpretability in black-box models, and persistent privacy and security concerns. Looking forward, we envision a future generative medicine ecosystem built on three pillars: multi-omic holographic Digital Twins integrating imaging with genomic and clinical data, enhanced human-machine collaboration paradigms, and comprehensive regulatory frameworks. This review provides essential insights for researchers and clinicians working to harness generative AI’s potential while addressing the fundamental challenges required for its responsible clinical deployment.

Ähnliche Arbeiten