Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Principled Confidence Estimation for Deep Computed Tomography
Matteo Gätzner, Johannes Kirschner
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use mixture of K predictions instead of their mean (ablation) | Diffusion / U-Net Ensemble · Sparse-view CT, three datasets | confidence coefficient beta_t: gap increases at higher intensity | Improved |
| use mixture instead of mean in dense-view setting (ablation) | Diffusion / U-Net Ensemble · Dense-view CT | confidence coefficient beta_t: difference <1% | Improved |
| use diffusion model as predictor instead of MLE | sequential likelihood mixing · Lamino (total intensity 1e9) | confidence gap: 2.3 x 10^5 -> 4.4 x 10^4 | Improved |
| use diffusion model as predictor instead of FBP | sequential likelihood mixing · Lamino (total intensity 1e9) | confidence gap: 1.6 x 10^6 -> 4.4 x 10^4 | Improved |
| use diffusion instead of U-Net for reconstruction | reconstruction · three CT datasets | PSNR: approximately 2 dB | Improved |
| use diffusion C_t-boundary sampling instead of worst-case optimization | pixel-wise uncertainty intervals · three CT datasets | interval width at high coverage: coverage near 100% -> high coverage with tighter intervals | Improved |
| use U-Net bootstrap instead of FBP bootstrap | pixel-wise intervals · three CT datasets | pixel coverage: low coverage vs overly conservative | Hurt |
| flag diffusion samples outside C_t as hallucinations | hallucination detection · Diffusion samples | PSNR: plausible samples have significantly higher PSNR | Improved |
Closest papers by meaning
- Efficient Noise Calculation in Deep Learning-based MRI Reconstructions (ICML 2025 · similarity 0.87)
- Provably Robust Score-Based Diffusion Posterior Sampling for Plug-and-Play Image Reconstruction (NeurIPS 2024 · similarity 0.85)
- Sparse Bayesian Deep Learning for Cross Domain Medical Image Reconstruction (AAAI 2024 · similarity 0.85)
- UGoDIT: Unsupervised Group Deep Image Prior Via Transferable Weights (NeurIPS 2025 · similarity 0.85)
- Towards Prospective Medical Image Reconstruction via Knowledge-Informed Dynamic Optimal Transport (NeurIPS 2025 · similarity 0.84)
- SGD Jittering: A Training Strategy for Robust and Accurate Model-Based Architectures (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
