←Back to feed
🧠 AI⚪ NeutralImportance 4/10
Prompting with the human-touch: evaluating model-sensitivity of foundation models for musculoskeletal CT segmentation
arXiv – CS AI|Caroline Magg, Maaike A. ter Wee, Johannes G. G. Dobbe, Geert J. Streekstra, Leendert Blankevoort, Clara I. S\'anchez, Hoel Kervadec|
🤖AI Summary
Researchers evaluated 11 promptable foundation models for medical CT image segmentation across bone and implant identification tasks. The study found significant performance variations between models and strategies, with all models showing sensitivity to human prompt variations, suggesting current benchmarks may overestimate real-world performance.
Key Takeaways
- →Segmentation performance varies significantly between foundation models and prompting strategies in medical imaging applications.
- →SAM and SAM2.1 are Pareto-optimal for 2D tasks, while nnInteractive and Med-SAM2 excel in 3D segmentation.
- →Human prompts consistently underperformed compared to ideal prompts extracted from reference labels.
- →All tested models demonstrated sensitivity to prompt variations, with limited robustness in inter-rater settings.
- →Model selection for human-driven medical segmentation remains challenging despite advances in foundation model technology.
#foundation-models#medical-imaging#segmentation#sam#ct-scans#human-prompts#model-evaluation#healthcare-ai
Read Original →via arXiv – CS AI
Act on this with AI
Stay ahead of the market.
Connect your wallet to an AI agent. It reads balances, proposes swaps and bridges across 15 chains — you keep full control of your keys.
Related Articles