Disentangled Learning Improves Implicit Neural Representations for Medical Reconstruction
This work addresses the inefficiency and quality limitations of current INRs for medical image reconstruction, offering a practical solution that requires no high-quality pre-acquired images.
DisINR introduces a disentangled learning framework for implicit neural representations that separates shared and subject-specific features, enabling efficient pre-training from raw measurements and test-time adaptation without forgetting. It achieves state-of-the-art reconstruction accuracy and efficiency across three medical imaging tasks.
Implicit neural representations (INRs) have emerged as a powerful paradigm for medical imaging via physics-informed unsupervised learning. Classical INRs optimize an entire network from scratch for each subject, leading to inefficient training and suboptimal imaging quality. Recent initialization-based approaches attempt to inject population priors into pre-trained networks, yet they rely on high-quality images and often suffer from catastrophic forgetting during fine-tuning. We present DisINR, a novel INR framework that explicitly disentangles shared and subject-specific representations. DisINR introduces a shared encoder-decoder pair and subject-specific encoders, whose features are jointly decoded for image reconstruction. By integrating differentiable forward models, it pre-trains the shared modules directly from limited raw measurements, removing the need for pre-acquired high-quality images. During test-time adaptation, only the subject-specific encoder is optimized, while the shared pair remains frozen, effectively preserving learned priors. Extensive evaluations on three representative medical imaging tasks show that DisINR significantly outperforms state-of-the-art INRs in both reconstruction accuracy and efficiency.