Deep learning reconstruction has rapidly moved from research curiosity to clinical reality, promising faster scans, improved image quality, and entirely new approaches to MRI acquisition. But behind the impressive vendor demonstrations and glossy marketing lies an important question: what is the technology doing and what are the risks, limitations, and opportunities for clinical MRI?

This talk will provide a practical and radiographer-focused overview of AI-based MR image reconstruction, covering the evolution from conventional Fourier and parallel imaging methods through compressed sensing to modern deep learning approaches.

Real-world examples from accelerated neuro, body, and musculoskeletal imaging will be presented, highlighting where AI reconstruction genuinely improves workflow and image quality and where caution is still required. The presentation will conclude with a discussion of whether AI reconstruction represents a fundamental shift in MRI physics or simply the latest chapter in a long history of increasingly sophisticated image processing.