Novel artificial intelligence-based hypodensity detection tool improves clinician identification of hypodensity on non-contrast computed tomography in stroke patients

• DOI: 10.3389/fneur.2024.1359775
• Publication Date: 2024-02-15T05:28:39Z
• Authors: Angela Dos Santos; Milanka Visser; Longting Lin; Andrew Bivard; Leonid Churilov; Mark William Parsons
• Keywords: Dice; Sørensen–Dice coefficient; Gold standard (test); Stroke

Introduction In acute stroke, identifying early changes (parenchymal hypodensity) on non-contrast CT (NCCT) can be challenging. We aimed to identify whether the accuracy of clinicians in detecting hypodensity ischaemic stroke patients a is improved with use an Artificial Intelligence (AI) based, automated detection algorithm (HDT) using MRI-DWI as gold standard. Methods The study employed case-crossover within-clinician design, where 32 were tasked lesions NCCT scans for five priori selected patient cases, before and after viewing AI-based HDT. DICE similarity coefficient (DICE score) was primary measure accuracy. Statistical analysis compared scores without HDT mixed-effects linear regression, individual nested random effects. Results had mean score 0.62 across all scans. Clinicians’ overall 0.33 (SD 0.31) implementation 0.40 0.27) implementation. associated increase 0.07 (95% CI: 0.02–0.11, p = 0.003) accounting scan clinician For small lesions, achieved 0.08 0.02, 0.13, 0.004) following use. subgroup 15 trainees, [mean difference 0.09 0.03, 0.14, 0.004)]. Discussion has potential enhance diagnosis, especially smaller notably less experienced clinicians.