ABSTRACT NUMBER: ESOC2026A259 COVERT SENTINEL INFARCTS ON NONCONTRAST CT: A HIGH-RISK CLINICAL SYNDROME PREDICTING STROKE ETIOLOGY AND OUTCOME

Abstract Background and aims Non-lacunar covert brain infarcts (CBIs) on noncontrast CT (NCCT) are frequently overlooked due to subtle hypoattenuation and unclear clinical implication, yet they often indicate high-risk etiologies such as large-artery atherosclerosis (LAA) or cardioembolism (CE). We investigated the clinical-etiological significance of non-lacunar CBIs and developed a deep learning–based model for automated detection. Methods We analyzed 628 patients with first-ever symptomatic ischemic stroke from a retrospective multicenter registry. We compared baseline characteristics, stroke etiology, and 3-month modified Rankin Scale (mRS) scores between patients with and without non-lacunar CBIs. Non-lacunar CBIs were subtyped as LAA- or CE-related to evaluate etiological concordance with the index stroke. A deep learning model was developed using a separate cohort (n=758) and validated in three independent external cohorts (n=1,680). Results Non-lacunar CBIs were identified in 13.5% (85/628) of patients. Their presence was associated with older age (75 vs 70 years; p=0.014), atrial fibrillation (34.1% vs 17.5%; p=0.002) and arterial occlusion (14.6% vs 7.0%; p=0.004). Etiological concordance was substantial: 48.5% of LAA-related CBIs matched a symptomatic LAA stroke, and 46.2% of CE-related CBIs matched a symptomatic CE stroke. Patients with non-lacunar CBIs had worse 3-month outcomes (mRS<3: 51.8% vs. 73.1%; adjusted p=0.002) and impaired recovery trajectories (p<0.001). The detection model achieved sensitivities 0.722–0.755 and specificities 0.797–0.932 across validation cohorts. Conclusions Non-lacunar CBIs on NCCT serve as sentinel markers showing high rate of etiological concordance with future devastating strokes and associated with poorer outcomes. An NCCT-based deep learning model shows robust generalization, enabling early identification and etiologic work-up. Conflict of interest WS Ryu, M Lee, D Kim and G Park are employees of JLK Inc., Seoul, Republic of Korea. SH Yeom is a former employee of the same company. Figure 1 - belongs to Methods Figure 1 - belongs to Results