A new medical imaging algorithm generates full 3D MRI scans up to 1000 times faster than current AI models.
April 25, 2026
Original Paper
WFM: 3D Wavelet Flow Matching for Ultrafast Multi-Modal MRI Synthesis
arXiv · 2604.21146
The Takeaway
MRI reconstruction usually requires minutes of processing time to transform raw data into a visible image. This new system starts with structured mathematical patterns called wavelets instead of random noise to skip the most time consuming steps. It cuts the generation time from 160 seconds down to a mere 0.16 seconds. This massive speed boost allows doctors to see high quality synthetic scans in real time during a patient visit. Clinicians could potentially use this to instantly visualize different types of tissue contrast without needing to perform multiple long scans. It marks a shift from waiting for results to seeing them the moment the scan begins.
From the abstract
Diffusion models have achieved remarkable quality in multi-modal MRI synthesis, but their computational cost (hundreds of sampling steps and separate models per modality) limits clinical deployment. We observe that this inefficiency stems from an unnecessary starting point: diffusion begins from pure noise, discarding the structural information already present in available MRI sequences. We propose WFM (Wavelet Flow Matching), which instead learns a direct flow from an informed prior, the mean o