Spatial consistency reveals uncertainty in black-box clinical vision models
SpatialUQ: Post-Hoc Uncertainty Quantification from Spatial Consistency in Black-Box Vision Models
Computer Vision and Pattern RecognitionArtificial IntelligenceMachine Learning
Summary
Clinical AI models often give predictions without showing how sure they are, and sometimes their internals or training data can’t be accessed during use. The authors introduce SpatialUQ, a method that measures uncertainty after the fact by comparing predictions from the whole image to predictions from parts of the image. If the predictions change a lot depending on the part of the image looked at, the model’s confidence might be low. Their method works well on chest X-ray data, is faster than some existing methods, and works best on diffuse problems rather than tiny features like nodules.
What this means in practice
- •For medical imaging software teams: Detect unreliable predictions from fixed black-box clinical vision AI models at inference time without needing retraining or ground truth.
- •For industrial quality assurance teams: Assess spatial consistency of AI model outputs in visual inspection systems to identify uncertain predictions without modifying existing models.
Authors
Md Kawsher Mahbub, Milon Biswas, Mirza Niaz Morshed, Wei Yu
Abstract
Clinical vision models are often deployed as frozen black boxes with no access to internals, retraining, or ground truth at inference time. We introduce \textbf{SpatialUQ}, a post-hoc uncertainty method using only output probabilities. It measures the Jensen-Shannon divergence between the global prediction and the mean of five fixed spatial crops in six deterministic forward passes. The premise is simple, trustworthy predictions are spatially consistent. On NIH ChestX-ray14 (DenseNet-121, $N{=}25{,}596$), our Multicrop Uncertainty Score (MUS) reaches $0.784$ failure-detection AUC versus $0.664$ for MC-Dropout ($p{<}10^{-6}$) at one-fifth the compute, with native calibration ($\text{SCE}{=}0.049$ vs.\ $0.127$ for $\ell_1$), the best-calibrated among methods above 0.78 AUC. A supervised fusion of MUS with entropy, confidence, and $\ell_1$ reaches $0.832$, outperforming a five-member ensemble ($0.813$). MUS scales with model quality, reaching $0.899$ with BiomedCLIP ($ρ= 0.846$), while this relationship remains meaningful in-distribution ($ρ= 0.523$) but breaks down under severe distribution shift (VinBigData, $ρ= 0.027$). MUS is well-suited to diffuse findings but is less dependable for small focal lesions such as nodules. Code and experimental materials are publicly available at https://huggingface.co/datasets/kawsher11/SpatialUQ.