A Deep Learning Model to Detect Acute MCA Occlusion on High-Resolution Noncontrast Head CT.

BACKGROUND AND PURPOSE: Stroke is a leading cause of morbidity and mortality around the world, with LVO having a disproportionate impact on patient outcomes. Our aim was to assess the feasibility and accuracy of a deep learning (DL) model to identify acute MCA occlusion by using high-resolution NCCT imaging data.

MATERIALS AND

METHODS

In this study, a total of 4648 consecutive examinations (July 2021 to December 2023) were retrospectively used for model training and validation, while an additional 1011 consecutive examinations (January 2024 to August 2024) were used for independent testing. By means of high-resolution NCCT acquired at a 1.0-mm slice thickness or less, MCA thrombus was labeled by using same-day CTA as ground truth. A 3D DL model was trained for per-voxel thrombus segmentation, with the sum of positive voxels used to estimate likelihood of acute MCA occlusion.

RESULTS

For detection of MCA M1 segment acute occlusion, the model yielded an area under the receiver operator curve (AUROC) of 0.952 [0.904-1.00], accuracy of 93.6% [88.1%-98.2%], sensitivity of 90.9% [83.1%-100%], and specificity of 93.6% [88.0%-98.3%]. Inclusion of M2 segment occlusions reduced performance only slightly, yielding an AUROC of 0.884 [0.825-0.942], accuracy of 93.2% [85.1%-97.2%], sensitivity of 77.4% [69.3%-92.2%], and specificity of 93.6% [85.1%-97.8%].

CONCLUSIONS

A DL model can detect acute MCA occlusion from high-resolution NCCT with accuracy approaching that of CTA. By means of this tool, most candidate thrombectomy patients may be identified with NCCT alone, possibly aiding stroke triage in settings that lack CTA or are otherwise resource-constrained.