Deep Learning–Enhanced Non-Contrast CT Lifts ASPECTS Agreement in Acute LVO Stroke

A DL-generated high-contrast thin-slice CT (HCCT) from plain NCCT pushed inter-rater ASPECTS correlation from 0.72 to 0.94 (p<0.001) in 303 LVO stroke patients across 4 centers — outperforming standard NCCT for early ischemic change detection and favorable-outcome prediction.

• Retrospective 4-center study (n=303; mean age 67.2 y; 2020–2024); two-stage framework applied DL super-resolution to NCCT to generate HCCT, benchmarked against Tmax >6 s perfusion maps as reference standard.
• ICC for HCCT-assisted ASPECTS was 0.85 (95% CI 0.80–0.89); ischemic volume correlated strongly with Tmax for both readers (r=0.83, p<0.001); ASPECTS ≥6 on HCCT showed strongest association with favorable outcome (OR 2.87, 95% CI 1.97–5.21), exceeding conventional NCCT.
• Key limitation: retrospective design with no prospective or external-site validation of the DL pipeline; real-world generalizability across scanner vendors and acquisition protocols remains unproven.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.