Evaluate the efficacy of AI-generated net water uptake (NWU) maps in early detection of ischemic stroke and prediction of lesion reversal.

Non-contrast computed tomography (NCCT) is typically the initial imaging study for suspected acute ischemic stroke but lacks sensitivity for detecting early ischemia. Net water uptake (NWU) measurement can characterize the ischemic core but is limited by the need CT perfusion. This study introduces an innovative artificial intelligence (AI) method to generate high-resolution NWU maps from NCCT and demonstrates its utility in visualizing potentially reversible acute ischemic stroke.

From a mechanical thrombectomy (MT) registry, patients with TICI 2b-3 recanalization were included. Z-score and NWU maps were derived from AI-denoised NCCT scans and visible ischemic tissue was labeled. For each lesion, reversal after intervention was determined and a NWU value was calculated by the difference in median lesion NWU with contralateral healthy tissue. The association of NWU with lesion reversal was assessed through univariate and multivariate analysis.

This study analyzed 116 distinct ischemic lesions from 52 patients. 29 lesions became undetectable on post-intervention imaging and 81 of the lesions were only detectable with the addition of NWU map compared to NCCT alone. In reversed lesions, NWU values were significantly lower (6.95% vs. 8.64%, p = 0.03) with the highest NWU observed being 15.7%. In multivariate analysis, higher NWU values were negatively associated with lesion reversal (Odds Ratio: 0.04 [0.00 – 0.66]), when adjusted for tPA administration and time delay between NCCT and MT (c-index: 0.66). Shapley analysis of the random forest classifier found that NWU was the most significant predictive feature for lesion reversal.

High-resolution NWU maps significantly improve visualization of early ischemia that is challenging to detect on NCCT. Lesion-specific NWU values suggest reversibility of ischemic lesions at higher values than previously reported. Validation in large patient cohorts is ongoing.