Principled Confidence Estimation for Deep Computed Tomography

We present a principled framework for confidence estimation in computed tomography (CT) reconstruction. Based on the sequential likelihood mixing framework (Kirschner et al., 2025), we establish confidence regions with theoretical coverage guarantees for deep-learning-based CT reconstructions. We consider a realistic forward model following the Beer-Lambert law, i.e., a log-linear forward model with Poisson noise, closely reflecting clinical and scientific imaging conditions. The framework is general and applies to both classical reconstruction algorithms and deep learning methods alike, including U-Nets, U-Net ensembles, and generative Diffusion models. Empirically, we demonstrate that deep reconstruction methods yield substantially tighter confidence regions than classical reconstructions, without sacrificing theoretical coverage guarantees. Our approach allows the detection of hallucinations in reconstructed images and provides interpretable visualizations of confidence regions. This establishes deep models not only as powerful estimators, but also as reliable tools for uncertainty-aware medical imaging.

The goal of computed tomography (CT) is to produce a 3-dimensional reconstruction of the inside of a body or object of interest from many X-ray measurements taken at different angles. Modern AI systems can produce strikingly sharp reconstruction from fewer measurement — which means less radiation — but they can also create details, known as hallucinations, that look real even when the measurements do not support them. This makes it important to know how much confidence we should place in each reconstruction. We developed a method that attaches a rigorous certificate of confidence to any CT reconstruction. Instead of a single picture, it maps out the full range of images that are genuinely consistent with the measurements, with a mathematical guarantee — for example 95% — that the range covers the true image. The method works with both traditional and AI-based reconstruction, accounts for the real physics of how X-rays pass through matter, and sharpens automatically as more measurements arrive; the better the reconstruction, the tighter and more useful the uncertainty quantification. This lets practitioners see how much to trust each part of an image, automatically flag AI hallucinations, and safely reduce radiation by stopping a scan once enough has been learned.