Related Experiment Video
Updated: Aug 6, 2026
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Automated Midline Shift Quantification on Noncontrast CT Across Intracranial Pathologies: A Multicenter Validation
Ish A Talati1, Sarah J Snyder1, Hafez D Haerian1
1From the Department of Radiology (I.A.T., J.J.H.), Stanford University School of Medicine, CA; Temple Medical School (S.J.S.), Philadelphia, PA; Department of Radiology (H.D.H.), Carle Foundation Hospital, Urbana, IL; Department of Radiology (J.M.H.), University of Colorado Medical School, Aurora, CO and Boulder Statistics (K.C.), Boulder, CO.
Background And Purpose:
Midline Shift (MLS) is caused by intracranial pathology and may result in fatal brain herniation. MLS measurement on non-contrast CT (NCCT) is time-consuming and subject to inter-rater variability. We evaluated the performance of Rapid MLS (iSchemaView, Inc.), an automated software tool for MLS quantification on NCCT.
Materials And Methods:
We performed a retrospective multicenter cohort study of adult patients with intracranial pathology and MLS. NCCT were collected from 13 centers, and all images had to be free of significant artifact to ensure accurate MLS measurement. MLS was manually measured by three independent neuroradiologists and the mean measurement was used as the gold standard. Reference standard MLS measurements were compared with automated MLS measurements from Rapid MLS. The primary outcome was non-inferiority of Rapid MLS to the average pairwise mean absolute error (MAE) among expert readers (one-sided α=0.025). Secondary analyses were agreement on Passing-Bablok regression and Bland-Altman analyses.
Results:
153 patients with NCCT met inclusion criteria. Mean patient age was 68.7 ± 16.3 years, and 89 patients (58%) were female. The cohort included a range of intracranial pathologies, such as intra- and extra-axial hemorrhage, and cerebral edema. MLS values ranged from 0-20.5 mm. The MAE of Rapid MLS (0.8 mm; 95% CI: 0.7-1.0 mm) was non-inferior to the average pairwise MAE among expert neuroradiologists (0.9 mm; 95% CI: 0.8-1.0 mm; p<0.0001). Passing-Bablok regression showed close agreement (intercept 0.29; slope 0.9). Bland-Altman analysis revealed a small mean difference of 0.2 mm (95% CI: 0.0-0.4 mm) between Rapid MLS and expert measurements. Rapid MLS tended to underestimate MLS values above 10 mm.
Conclusions:
Rapid MLS provided automated MLS measurements on NCCT with accuracy comparable to that of expert neuroradiologists in this multicenter cohort. These findings support its potential use as an adjunct for rapid and reproducible MLS quantification in acute neuroimaging workflows.
