Explainable Knowledge-Guided Algorithm for Contrast Extravasation Detection on Computed Tomography

To develop an explainable, knowledge-guided framework for automated detection of contrast media extravasation from sequential computed tomography (CT) images and to evaluate its potential to accelerate time-critical trauma triage while maintaining clinically acceptable sensitivity.

A mathematical framework was formulated to explicitly encode three expert-derived diagnostic rules: 1) progressive increase of contrast outside anatomically plausible vessels, 2) appearance of contrast in non-vascular regions, and 3) localized irregularity of vessel caliber. Sequential two-dimensional CT slices were analyzed using a 2.5D formulation integrating temporal intensity evolution, anatomical plausibility, vessel morphology, and inter-slice continuity. The model outputs a confidence score and a binary alert. Model parameters and decision thresholds were initialized using a single representative clinical case guided by expert interpretation. Performance was evaluated against senior emergency surgeon assessment, emphasizing sensitivity and time-to-decision.

The proposed framework achieved clinically acceptable sensitivity for detection of contrast extravasation while substantially reducing time-to-decision relative to manual review. Early-trigger analysis demonstrated that positive cases were identified within the initial portion of the CT volume, supporting rapid screening and prioritization in emergency workflows.

This study demonstrates the feasibility of translating expert clinical reasoning into an interpretable computational model for time-critical imaging tasks. The knowledge-guided design enables rapid automated screening while preserving transparency and clinician oversight. The framework shows promise as a decision-support tool for accelerating trauma triage, with future work focused on prospective validation and broader multi-center evaluation.

The proposed knowledge-guided algorithm enables rapid extravasation alerts on trauma CT, supporting earlier triage and prioritization for angiography or surgery within existing emergency imaging workflows.