Table 1 Patient-level performance for detection and localization of vessel occlusions in the FAST 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 = 52)/Controls (n = 274)

0.90 [0.84, 0.94]

45/52; 0.87 [0.77, 0.95]

212/274; 0.77 [0.72, 0.82]

45/107; 0.42 [0.32, 0.51]

212/219; 0.97 [0.94, 0.99]

VO + HGS (n = 79)/Controls (n = 247)

0.92 [0.88, 0.96]

69/79; 0.87 [0.79, 0.94]

209/247; 0.85 [0.80, 0.90]

69/107; 0.64 [0.55, 0.74]

209/219; 0.95 [0.92, 0.98]

Class: VO

Anterior Circulation

LVO (n = 23)

0.89 [0.80, 0.97]

20/23; 0.87 [0.73, 1.0]

-

20/82; 0.24 [0.15, 0.34]

212/215; 0.99 [0.97, 1.0]

MeVO (n = 14)

0.93 [0.87, 0.97]

14/14; 1.0 [1.0, 1.0]

-

14/76; 0.18 [0.1, 0.27]

212/212; 1.0 [1.0, 1.0]

Posterior Circulation

VO (n = 10)

0.82 [0.65, 0.93]

6/10; 0.60 [0.25, 0.89]

-

6/66: 0.09 [0.03, 0.16]

212/216; 0.98 [0.96, 1.00]

Class: VO + HGS

Anterior Circulation

LVO (n = 32)

0.94 [0.89, 0.98]

30/32; 0.94 [0.85, 1.0]

-

30/68; 0.44 [0.33, 0.56]

209/211; 0.99 [0.98, 1.0]

MeVO (n = 20)

0.96 [0.94, 0.98]

20/20; 1.00 [1.0, 1.0]

-

20/58; 0.34 [0.23, 0.46]

209/209; 1.00 [1.0, 1.0]

Posterior Circulation

 VO (n = 13)

0.79 [0.65, 0.91]

6/13; 0.46 [0.2, 0.75]

-

6/44; 0.14 [0.04, 0.24]

209/216; 0.97 [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%).