Abstract TP424: Automated Detection of Lacunes in a Cohort of Patients Presenting With Transient Ischemic Attack or Acute Ischemic Stroke

Abstract

Introduction: Cerebral Small Vessel Disease (CSVD) impacts the functional outcome of acute ischemic stroke (AIS) patients. The majority of the available tools to quantify its burden relies on visual categorical scales, which is labor intensive and subject to inter- and intra-reader variability. Fully automated tools are rare and mostly focus on white matter hyperintensity (WMH) as a surrogate for CSVD. We aimed to develop a novel software that simulates the clinicians’ rational to detect lacunes on clinical MRI scans of patients presenting with acute AIS or transient ischemic attack (TIA). Methods: Patients presenting with symptoms of acute AIS or TIA were prospectively recruited. Lacunes were scored on the first brain MRI collected within 24 hours of hospital admission by a board-certified neuro-intensivist/vascular neurologist according to the Neuroimaging Standards for Research into Small Vessel Disease (STRIVE) criteria. Following standard skull stripping and co-registration, automated software was developed in Matlab by calculating maximal intensity difference of co-registered voxels on FLAIR and T2 MRI sequences. Cerebrospinal fluid was removed by seed-based growing approach and lacunes were subsequently detected based on their size (3-15 mm) and morphological features. Results: 30 subjects were included (age 61.6±16.1, 30% females); 6 of which had TIA and 23 had AIS (10 acute lacunar stroke, 3 cardioembolic, 10 cryptogenic and 1 stroke of other determined etiology). There were 24 lacunes detected in 12 subjects by the human reader. The automated system accurately identified subjects with lacunes in 91.6% of the cases with 75% specificity. Negative predictive value for subjects without lacunes was 93.7%. At the lesion level, the paradigm identified 62.5% of all lacunes in all subjects with positive predictive value of 78.9%; 67.0% of the missed lacunes were in close proximity to the ventricles. Conclusions: Automated identification of lacunes is feasible on clinical MRI scans of patients with acute AIS or TIA. Lacunes closer to the ventricles are challenging and may require a separate approach.