An Explainable Agentic AI Framework for Uncertainty-Aware and Abstention-Enabled Acute Ischemic Stroke Imaging Decisions

TL;DR

The paper tackles safety and trust in AI-assisted acute ischemic stroke imaging by moving beyond black-box predictions toward uncertainty-aware, abstention-enabled decision support. It proposes a modular agentic AI framework with perception, uncertainty estimation, and decision agents, complemented by explainability modules and governed by a slice-level uncertainty threshold $\tau$. Qualitative analyses demonstrate that abstention naturally concentrates in diagnostically ambiguous or low-information regions, while explanations accompany both predictions and deferrals. The work emphasizes clinician alignment and safety as core design principles, arguing that agentic control and selective abstention are essential for trustworthy medical imaging AI in high-risk emergency settings.

Abstract

Artificial intelligence models have shown strong potential in acute ischemic stroke imaging, particularly for lesion detection and segmentation using computed tomography and magnetic resonance imaging. However, most existing approaches operate as black box predictors, producing deterministic outputs without explicit uncertainty awareness or structured mechanisms to abstain under ambiguous conditions. This limitation raises serious safety and trust concerns in high risk emergency radiology settings. In this paper, we propose an explainable agentic AI framework for uncertainty aware and abstention enabled decision support in acute ischemic stroke imaging. The framework follows a modular agentic pipeline in which a perception agent performs lesion aware image analysis, an uncertainty estimation agent computes slice level predictive reliability, and a decision agent determines whether to issue a prediction or abstain based on predefined uncertainty thresholds. Unlike prior stroke imaging systems that primarily focus on improving segmentation or classification accuracy, the proposed framework explicitly prioritizes clinical safety, transparency, and clinician aligned decision behavior. Qualitative and case based analyses across representative stroke imaging scenarios demonstrate that uncertainty driven abstention naturally emerges in diagnostically ambiguous regions and low information slices. The framework further integrates visual explanation mechanisms to support both predictive and abstention decisions, addressing a key limitation of existing uncertainty aware medical imaging systems. Rather than introducing a new performance benchmark, this work presents agentic control, uncertainty awareness, and selective abstention as essential design principles for developing safe and trustworthy medical imaging AI systems.

Figures (4)

• Figure 1: The proposed explainable agentic AI framework for uncertainty-aware acute stroke imaging employs a hierarchical top-down design. This design distinctly separates perception, uncertainty estimation, and safety-aware decision-making processes. Such a structure facilitates abstention in situations of high epistemic uncertainty and allows for clinician-in-the-loop intervention.
• Figure 2: Slice-wise uncertainty profile across axial brain slices. Elevated uncertainty is observed in ambiguous or low-information regions, motivating abstention in safety-critical decision-making.
• Figure 3: The visualization of decision outcomes influenced by uncertainty across representative image slices. Slices characterized by low uncertainty result in confident predictions, whereas those with high uncertainty necessitate abstention and referral to a clinician.
• Figure 4: visualizing decision outcomes influenced by uncertainty on axial brain slices, we gain a deeper understanding of system behavior. In diagnostically clear sections, confident predictions are made, whereas in adjacent or ambiguous areas, abstention occurs, mirroring clinically appropriate deferral behavior.