CogGen: Cognitive-Load-Inspired Fully Unsupervised Deep Generative Modeling for Compressively Sampled MRI Reconstruction

CogGen is a novel fully unsupervised deep generative modeling framework designed for compressively sampled magnetic resonance imaging (CS-MRI) reconstruction. It employs a self-paced curriculum learning strategy that adapts k-space measurement participation based on a dual-threshold weighting criterion, improving convergence and reducing noise amplification. Numerical experiments show that CogGen, implemented as CogGen-DIP and CogGen-INR, outperforms existing CS-MRI reconstruction methods, making it a significant advancement for practitioners seeking more efficient and accurate MRI reconstruction techniques.