Table 1 Algorithmic classification performance of stroke patterns (above) and vascular territories (below)

From: DeepISLES: a clinically validated ischemic stroke segmentation model from the ISLES'22 challenge

 

Stroke pattern

 
 

SVI (N=62)

SI based on micro-occlusions (N = 48)

SVI with accompanying SI (N = 38)

All stroke (N = 148)

Team

F1 score

Balanced Accuracy

SEALS

87.6

75.6

68.8

78.1

NVAUTO

88.1

78.6

68.1

78.9

SWAN

85.0

75.9

68.2

76.2

DeepISLES

87.6

91.8

81.6

86.9

 

Vascular territory

 
 

MCA (N = 97)

PCA (N = 23)

ACA (N = 4)

Pons/Medula (N = 4)

Cerebellum (N = 20)

All stroke (N = 148)

Team

F1 score

Balanced Accuracy

SEALS

97.9

93.3

88.9

80.0

97.6

97.4

NVAUTO

97.9

95.7

80.0

88.9

97.4

97.3

SWAN

96.8

93.3

66.7

100.0

97.6

92.2

DeepISLES

98.4

93.3

100.0

88.9

97.6

97.6

  1. DeepISLES is notably superior to any individual solution in identifying the stroke pattern and the vascular territory. All metrics are reported in percentage values. The best results are highlighted in bold. Source data are provided as a Source Data file.
  2. SVI single vessel infarct, SI scattered infarcts, MCA middle cerebral artery, ACA anterior cerebral artery, PCA posterior cerebral artery.