AI fixes 95.8% of failed spinal-cord labels: nnU-Net rescues patient MRI data
bravo_abad · x · 2026-10-25
Brain MRIs often capture the top of the spinal cord, and mislabeled vertebral levels can turn real biological differences into processing errors.
- The established SCT tool mislabeled levels in 47% of scans from hereditary ataxia patients vs 28% of healthy controls — and discarding failures would disproportionately drop patient data
- Jarola et al. trained an nnU-Net segmentation model on visually verified labels (739 training, 185 validation)
- On the 698 scans the existing tool had mislabeled, the model labeled 669 correctly: 95.8%
- The team also built a deep-learning automated quality-control model to filter bad segmentations
More from AGI Musings
- AI agent books an impossible cardiologist appointment through Goa's broken portal — alysha_lobo · 2026-10-25
- There's no moat in code anymore: what's left is distribution, taste, community — paw_lean · 2026-10-25
- Cartoon skewers AI alignment: chained robots taught "human values" in a classroom — repligate · 2026-10-25
- Copywriters most AI-anxious at 33%, engineers near 24% — job worry survey ranks top 10 — oran_ge · 2026-10-25
- One year ago, entrepreneurs topped the 'least worried about AI' poll at 83% — oran_ge · 2026-10-25
- Reddit debate: AI removed the coding excuse, so why is nobody building million-dollar ideas? — Positive-Anywhere872 · 2026-10-25