AI-Based Automated Quantification of Arterial Stenosis in Head and Neck CT Angiography: A Comparison with Manual Measurements from Digital Subtraction Angiography and CT Angiography.

Abstract

Purpose: To evaluate the performance of an artificial intelligence (AI) algorithm for automated quantification of arterial stenosis in head and neck CT angiography (CTA).

Methods: Patients who received head and neck CTA and DSA between January 2019 and December 2021 in two centers were included. The quantitative performance of CerebralDoc per-lesion was evaluated through intraclass correlation coefficients (ICCs) and Bland-Altman analysis, comparing automated stenosis measurements and manual measurements across 0-100%, < 50%, ≥ 50% and ≥ 70% thresholds. Sensitivity analysis included linear and logistic regression, and subgroups analysis was performed to identify influencing factors.

Results: 287 patients with 1765 lesions were analyzed. ICCs between CerebralDoc and DSA for ≥ 50% and ≥ 70% stenosis were excellent (0.955, 0.922, respectively), for 0-100% stenosis was good (0.735), and for < 50% stenosis was poor (0.056). For ≥ 50% and ≥ 70% stenosis of CerebralDoc and CTA manual measurements versus DSA, ICCs were close (0.955 vs 0.994; 0.922 vs 0.986), and differences were small (0.258% vs -0.362%; 0.369% vs -0.199%). The sensitivity analysis revealed that specific locations (V1, V2, V3, V4) and slender vessels have large or remarkable differences ranging from 15.551% to 44.238%.

Conclusion: CerebralDoc exhibited excellent performance in automatically quantifying arterial stenosis of ≥ 50% and ≥ 70% in head and neck CTA. However, further research was needed to improve its performance for < 50% stenosis and to address differences in specific locations and slender vessels.