Artificial intelligence to accelerate MRI scans - Philips & LUMC winners in the...
summary
Philips & LUMC win the fastMRI challenge by submitting an adaptive compressed sensing network that fuses deep learning with Fourier/physics-based steps, uses residual blocks and plug‑in transforms, and is fine‑tuned per protocol. Reconstructions show dramatic artifact reduction versus undersampled inverses, clinically validated by radiologist evaluation, and claim acceleration up to a factor of eight. Emphasis on bespoke loss functions and hybrid implementation — simple blocks, iterative structure, and measurable diagnostic improvements.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:14AI in MRI acceleration
- 6:51AI for MRI performance
- 8:18MRI challenges and limits
- 9:22First MRI challenge
- 11:10Philips’ MRI win
- 12:14Hybrid AI for MRI
- 14:20Compressed sensing + deep learning
- 16:09Deep learning as approximation
- 17:54Model architecture choices
- 19:16Loss functions role
- 20:24Image quality beyond metrics
- 22:52Training and fine-tuning
- 24:35Results and clinical impact
- 27:05Key takeaways
- 28:08Q&A: Team dynamics
- 29:11Q&A: Clinical timeline
- 29:55Q&A: Future challenges
- 31:00Q&A: Radiologist loss design
related talks
50mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
42mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
49mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
43mBuilding AI That Learns and Adapts: A Case Study in MRI Diagnostics
44mBuilding Adaptive AI for MRI Diagnostics: A Continual Learning Case Study
10mSeeing the World With AI: How Deep Learning Enables a New Era of Imaging Technologie
36mAI in Healthcare
34mWhy AI in the Medical Field Both Sucks and Rocks...