From Time to Tissue: AI-Driven Net Water Uptake Mapping Enables Early Detection of Reversible Ischemia on Non-Contrast CT Scans

Abstract

<title>Abstract</title> Background: Non-contrast computed tomography (NCCT) is the most widely used initial imaging modality for acute ischemic stroke (AIS) triage, but its limited sensitivity to early ischemia changes constrains accurate lesion characterization. Net water uptake (NWU) has emerged as a promising biomarker for delineating the ischemic core, yet its application has been restricted by reliance on CT perfusion (CTP). This study introduces an innovative artificial intelligence (AI)-enabled method to generate high resolution, whole brain NWU maps and Z-score maps directly from NCCT scans, enabling the detection of both ischemic core and potentially salvageable tissue without the need for contrast agents or perfusion imaging. Methods: We retrospectively analyzed data from a mechanical thrombectomy (MT) stroke registry, including patients who achieved successful TICI 2b-3 recanalization. NWU and Z-score maps were derived from NCCT scans using AI denoising and bias correction. Visible ischemic tissue was labeled, and lesion fate following intervention was determined using follow-up MRI. NWU values were computed for each lesion and correlated with tissue reversibility post-reperfusion using univariate and multivariate logistic regression models. Results: This study analyzed 237 distinct ischemic lesions from 94 patients. Standard NCCT identified lesions in 35 of 94 patients, yielding a sensitivity of 37.2%. In contrast, 82 patients (87.2%) had lesions that were occult on standard NCCT but became detectable with the addition of AI-generated NWU maps. Post-intervention imaging demonstrated complete lesion reversal in 33 patients (35.1%). Reversed lesions exhibited significantly lower NWU values than non-reversed lesions (6.99% vs 10.14%, P&lt;.001), indicating a strong association between reduced water uptake and tissue viability. After adjusting NWU for CT to MT time delay and regional tissue-specific reversibility thresholds, we developed a significant predictor strongly associated with stroke lesion reversal (Odds Ratio, 3.20; 95% CI, 1.18–8.67; P=.02). Conclusions: Our study introduces an AI-driven method that generates high-resolution NWU and Z-score maps directly from standard NCCT scans, enabling visualization and quantitative assessment of ischemic tissue viability without the need for advanced imaging such as CTP or MRI. We found that lower NWU values are significantly associated with lesion reversibility following reperfusion therapy. This establishes NWU as a robust, imaging-derived biomarker for predicting tissue fate in acute stroke. By enabling individualized, physiology-driven assessment at the point of care, this technique supports a paradigm shift from rigid time-based treatment thresholds to a more nuanced, tissue-based triage framework for stroke management.

Ähnliche Arbeiten