BMC Neurology Неврология (клиническая)

Regional correspondence between visual Fazekas ratings and AI-derived white matter hyperintensity burden in acute ischemic stroke: a retrospective cross-sectional study

Qiaoqiao Liao, Yue Tang, Hailong Lian, Liping Wang, Jinping Liu
Springer Nature · 2026-11-17 · DOI 10.1186/s12883-026-05594-x

Аннотация

Abstract Background Visual Fazekas ratings assess white matter hyperintensity (WMH) severity, whereas artificial intelligence (AI)-derived segmentation provides continuous measurements. We quantified regional and total visual–AI correspondence, inter-reader reliability, and high-burden concordance in acute ischemic stroke. Methods We included 361 patients with acute ischemic stroke undergoing T2-weighted fluid-attenuated inversion recovery (T2-FLAIR) magnetic resonance imaging (MRI). Two readers independently assigned periventricular, deep, and total Fazekas scores. AI-derived WMH volumes were normalized to intracranial volume. Spearman correlations, HC3-robust linear trends, quadratic weighted κ, binary agreement, and a 2 × 3 threshold-sensitivity matrix were evaluated. Visual high burden was total Fazekas score ≥ 4; AI high burden was the cohort-specific upper third of total WMH/intracranial volume (1.20%). Results Median age was 67 years, and 258 patients (71.5%) were men. Reader 1 ratings correlated strongly with AI-derived periventricular WMH (ρ = 0.840, 95% CI 0.807–0.866) and total WMH (ρ = 0.880, 95% CI 0.851–0.902), and moderately with deep WMH (ρ = 0.588, 95% CI 0.514–0.656; all P < 0.001). Inter-reader reliability was excellent (quadratic weighted κ = 0.869–0.926). Under the primary thresholds, 326 of 361 patients agreed (90.3%), with κ = 0.781 (95% CI 0.708–0.847). Reader 2 replication was similar; across alternative thresholds, κ ranged from 0.462 to 0.801. Conclusions Visual Fazekas ratings showed strong correspondence with AI-derived WMH burden, especially for total and periventricular disease. Weaker deep-WMH correspondence and threshold-dependent discordance indicate that visual ratings and quantitative measurements are complementary rather than interchangeable.

Читать у издателя →