Table 1 Dataset details: the properties of the datasets that were used for model training and testing.

From: Development and clinical application of a deep learning model to identify acute infarct on magnetic resonance imaging

 

Training set

Validation set

Primary test set

Stroke code test sets

International test set

Classification

Segmentation

Classification

Segmentation

Classification

Segmentation

Training hospital

Non-training hospital

Number of studies

6657

377

725

34

792

62

381

247

171

Number of positive studies (%)

3314 (49.8%)

All

372 (51.3%)

All

384 (48.5%)

All

168 (44.1%)

128 (50.2%)

70 (40.9%)

Time period of studies

01/2004–05/2018

01/2004–05/2018

01/2007–05/2018

02/2007–05/2018

01/2007–05/2018

03/2007–05/2018

07/2018–01/2019

07/2018–12/2018

01/2017–07/2019

Number of studies on female patients (%)

3445 (51.8%)

176 (46.7%)

374 (51.6%)

17 (50.0%)

404 (51.0%)

26 (41.9%)

193 (50.7%)

129 (52.2%)

101 (59.1%)

Mean age in years ± standard deviation (range)

60.7 ± 18.0 (18–104)

68.1 ± 14.6 (18–102)

60.8 ± 17.7 (18–101)

67.4 ± 18.4 (26–96)

60.5 ± 18.4 (18–102)

68.2 ± 15.8 (26–99)

65.9 ± 16.5 (19–98)

67.6 ± 17.2 (22–97)

46.7 ± 21.1

(18–95)

Median infarct volume in mL (interquartile range; range)

6.42 (0.61–33.28; 0.02–333.06)

6.38 (1.42–36.03; 0.06–276.38)

5.57 (0.78–56.88; 0.03–308.78)

2.73 (0.47–12.98; 0.04–403.16)

6.12 (1.03–43.47; 0.10–442.80)

3.01 (0.75–14.54; 0.07–255.20)

Number of studies on GE scanners

5233

349

568

31

618

57

345

52

Unavailable

Number of studies on Siemens scanners

1424

28

157

3

174

5

36

196

Unavailable