Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evidential Perfusion Physics-Informed Neural Networks with Residual Uncertainty Quantification

Junhyeok Lee, Minseo Choi, Han Jang, Young Hun Jeon, Heeseong Eum, Joon Jang, Chul-Ho Sohn, Kyu Sung Choi

TL;DR

Evidential Perfusion Physics-Informed Neural Networks (EPPINN) is proposed, a framework that integrates evidential deep learning with physics-informed modeling to enable uncertainty-aware perfusion parameter estimation and attains the highest voxel-level and case-level infarct-core detection sensitivity.

Abstract

Physics-informed neural networks (PINNs) have shown promise in addressing the ill-posed deconvolution problem in computed tomography perfusion (CTP) imaging for acute ischemic stroke assessment. However, existing PINN-based approaches remain deterministic and do not quantify uncertainty associated with violations of physics constraints, limiting reliability assessment. We propose Evidential Perfusion Physics-Informed Neural Networks (EPPINN), a framework that integrates evidential deep learning with physics-informed modeling to enable uncertainty-aware perfusion parameter estimation. EPPINN models arterial input, tissue concentration, and perfusion parameters using coordinate-based networks, and places a Normal--Inverse--Gamma distribution over the physics residual to characterize voxel-wise aleatoric and epistemic uncertainty in physics consistency without requiring Bayesian sampling or ensemble inference. The framework further incorporates physiologically constrained parameterization and stabilization strategies to promote robust per-case optimization. We evaluate EPPINN on digital phantom data, the ISLES 2018 benchmark, and a clinical cohort. On the evaluated datasets, EPPINN achieves lower normalized mean absolute error than classical deconvolution and PINN baselines, particularly under sparse temporal sampling and low signal-to-noise conditions, while providing conservative uncertainty estimates with high empirical coverage. On clinical data, EPPINN attains the highest voxel-level and case-level infarct-core detection sensitivity. These results suggest that evidential physics-informed learning can improve both accuracy and reliability of CTP analysis for time-critical stroke assessment.

Evidential Perfusion Physics-Informed Neural Networks with Residual Uncertainty Quantification

TL;DR

Evidential Perfusion Physics-Informed Neural Networks (EPPINN) is proposed, a framework that integrates evidential deep learning with physics-informed modeling to enable uncertainty-aware perfusion parameter estimation and attains the highest voxel-level and case-level infarct-core detection sensitivity.

Abstract

Physics-informed neural networks (PINNs) have shown promise in addressing the ill-posed deconvolution problem in computed tomography perfusion (CTP) imaging for acute ischemic stroke assessment. However, existing PINN-based approaches remain deterministic and do not quantify uncertainty associated with violations of physics constraints, limiting reliability assessment. We propose Evidential Perfusion Physics-Informed Neural Networks (EPPINN), a framework that integrates evidential deep learning with physics-informed modeling to enable uncertainty-aware perfusion parameter estimation. EPPINN models arterial input, tissue concentration, and perfusion parameters using coordinate-based networks, and places a Normal--Inverse--Gamma distribution over the physics residual to characterize voxel-wise aleatoric and epistemic uncertainty in physics consistency without requiring Bayesian sampling or ensemble inference. The framework further incorporates physiologically constrained parameterization and stabilization strategies to promote robust per-case optimization. We evaluate EPPINN on digital phantom data, the ISLES 2018 benchmark, and a clinical cohort. On the evaluated datasets, EPPINN achieves lower normalized mean absolute error than classical deconvolution and PINN baselines, particularly under sparse temporal sampling and low signal-to-noise conditions, while providing conservative uncertainty estimates with high empirical coverage. On clinical data, EPPINN attains the highest voxel-level and case-level infarct-core detection sensitivity. These results suggest that evidential physics-informed learning can improve both accuracy and reliability of CTP analysis for time-critical stroke assessment.
Paper Structure (13 sections, 9 equations, 4 figures, 2 tables)

This paper contains 13 sections, 9 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. CTP Forward Model and Physics Constraints
  4. Architecture and Evidential Formulation
  5. Parameterization and Optimization Stability
  6. Delay and MTT Anti-Collapse Regularization
  7. Experiments
  8. Datasets
  9. Baselines and Implementation Details
  10. Digital Phantom Experiments
  11. Clinical Experiments
  12. Ablation Studies
  13. Conclusion

Figures (4)

  • Figure 1: Overview of EPPINN. Coordinate-based networks model AIF and tissue curves, and a parameter network predicts perfusion parameters and evidential residual uncertainty.
  • Figure 2: Digital phantom results under controlled noise and temporal sampling. Left: NMAE for $\mathrm{CBF}$/$\mathrm{CBV}$ across PSNR and $\delta t$. Right: empirical coverage of uncertainty intervals.
  • Figure 3: Qualitative perfusion maps on ISLES (left) and a clinical case (right): EPPINN, ReSPPINN, and boxNLR ($\mathrm{CBF}$/$\mathrm{CBV}$/$\mathrm{MTT}$/$T_{\max}$).
  • Figure 4: Training dynamics and residual uncertainty. Left: iterative $\mathrm{CBF}$ evolution across methods. Right: final $\mathrm{CBF}$ comparison between EPPINN and PINN (EPPINN without evidential loss), with aleatoric, epistemic, and total residual uncertainty maps.