Integrated Deep Learning-Based Intracranial Vessel Wall Imaging with DANTE Preparation: Feasibility and Technical...

Although DANTE preparation improves blood suppression in intracranial vessel wall imaging, its integration with deep learning-accelerated T1-SPACE and the resulting lumen-wall image-quality trade-off remain poorly characterized. We evaluated the feasibility and technical performance of integrating a delay alternating with nutation for tailored excitation (DANTE) preparation into a deep learning-accelerated, postcontrast T1-SPACE sequence for intracranial vessel wall imaging (IC-VWI).

In this retrospective, single-center study, 35 patients (22 women; mean age, 57.9 ± 17.1 years) underwent IC-VWI using postcontrast deep learning-T1-SPACE with (T1-SPACE_DL+DANTE) and without (T1-SPACE_DL) a DANTE preparation. Two neuroradiologists independently scored lumen and wall visualization across the arterial segments on a 4-point Likert scale (1: worst to 4: best) and graded venous flow artifacts along the MCA, perimesencephalic veins (PMV), deep cerebral veins (DCV), and cortical veins (CV). Intersequence comparisons used cumulative-link mixed-effects models (CLMMs); segments were additionally pooled and analyzed as proximal versus distal. Venous flow artifact scores were compared with paired Wilcoxon tests between sequences and percentage agreement between readers. Exploratory Bland-Altman analysis was also performed for both readers.

A total of 556 vessel segment pairs were analyzed. In CLMM analysis, T1-SPACE_DL+DANTE improved lumen scores versus T1-SPACE_DL (pooled OR 40.02; 95% CI 24.06-66.57; false discovery rate [FDR] P < .001) but reduced wall scores (pooled OR 0.11; 95% CI 0.08-0.14; FDR P < .001). By anatomic group, lumen ORs were 26.03 (proximal) and 91.93 (distal), and wall ORs were 0.12 (proximal) and 0.04 (distal) (all FDR P < .001). Venous flow artifacts improved across all analyzed subsites (P < .001). The ±1-point interreader concordance was near perfect across analyses. Bland-Altman plots showed negative lumen bias (favoring T1-SPACE_DL+DANTE) and positive wall bias (favoring T1-SPACE_DL) without consistent proportional bias.

Adding DANTE preparation to deep learning-accelerated IC-VWI was associated with fewer flow-related artifacts and a clearer depiction of the vessel lumen, which may support a more accurate assessment of intracranial vasculopathies and aneurysms. Potential gains were accompanied by a modest wall-visualization penalty, which is not unexpected with a flow-suppression pulse.