Table 2 Patient-level performance for detection and localization of vessel occlusions in the UKB cohort

From: Deep-learning based detection of vessel occlusions on CT-angiography in patients with suspected acute ischemic stroke

Cohort n = Num Patients

AUROC

Sensitivity

Specificity

PPV

NPV

Full Dataset

 VO (n = 80)/Controls (n = 243)

0.85 [0.79, 0.91]

65/80; 0.81 [0.71, 0.90]

196/243; 0.81 [0.75, 0.85]

65/112; 0.58 [0.49, 0.67]

196/211; 0.93 [0.89, 0.96]

 VO + HGS (n = 106)/Controls (n = 217)

0.88 [0.83, 0.93]

85/106; 0.80 [0.72, 0.87]

190/217; 0.88 [0.83, 0.92]

85/112; 0.76 [0.68, 0.83]

190/211; 0.90 [0.86, 0.94]

Class: VO

Anterior Circulation

 LVO (n = 29)

0.95 [0.88, 0.99]

28/29; 0.97 [0.89, 1.0]

-

28/75; 0.37 [0.27, 0.48]

196/197; 0.99 [0.98, 1.0]

 MeVO (n = 27)

0.79 [0.66, 0.90]

20/27; 0.74 [0.57, 0.89]

-

20/67; 0.30 [0.20, 0.41]

196/203; 0.97 [0.94, 0.99]

Posterior Circulation

 VO (n = 15)

0.79 [0.63, 0.92]

10/15; 0.67 [0.40, 0.90]

-

10/57; 0.18 [0.08, 0.28]

196/201; 0.98 [0.95, 0.99]

Class: VO + HGS

Anterior Circulation

 LVO (n = 41)

0.95 [0.89, 0.99]

37/41; 0.90 [0.81, 0.98]

-

37/64; 0.58 [0.45, 0.70]

190/194; 0.98 [0.96, 1.00]

 MeVO (n = 29)

0.85 [0.74, 0.95]

23/29; 0.79 [0.62, 0.94]

-

23/50; 0.46 [0.33, 0.60]

190/196; 0.97 [0.94, 0.99]

Posterior Circulation

 VO (n = 20)

0.83 [0.71, 0.93]

13/20; 0.65 [0.43, 0.86]

-

13/40; 0.33 [0.17, 0.47]

190/197; 0.96 [0.94, 0.99]

  1. Patients with multiple occlusions were ignored for the automated analysis in the per-class category.
  2. VO vessel occlusion, HGS high-grade stenosis (>70%).