Table 2 Performances of ResNet-50 in Data Set 1 and Data Set 2. Parentheses show 95% CIs.

From: Automatic classification of canine thoracic radiographs using deep learning

Test set

Radiographic finding

AUC

Sensitivity

Specificity

PLR

NLR

Data Set 1

Alveolar pattern

0.87 (0.78–0.97)

0.95 (0.64–1)

0.38 (0.31–0.45)

1.48 (1.2–1.8)

0.2(0.01–1.4)

Data Set 2

Alveolar pattern

0.89 (0.86–0.92)

0.95 (0.9–0.98)

0.52 (0.38–0.72)

1.99 (1.8–2.2)

0.095 (0.04–0.2)

Data Set 1

Bronchial pattern

0.78 (0.66–0.9)

0.95 (0.66–0.99)

0.092 (0.04–0.68)

1.02 (0.9–1.2)

0.78(0.1–0.54)

Data Set 2

Bronchial pattern

0.69 (0.61–0.76)

0.96 (0.86–0.99)

0.20 (0.17–0.24)

1.2 (1.1–1.3)

0.2 (0.05–0.8)

Data Set 1

Cardiomegaly

0.92 (0.88–0.97)

0.95 (0.86–1)

0.52 (0.43–0.6)

1.98 (1.7–2.3)

0.08 (0.02–0.3)

Data Set 2

Cardiomegaly

0.89 (0.86–0.92)

0.95 (0.91–0.98)

0.59 (0.54–0.63)

2.31 (2.1–2.6)

0.076 (0.03–0.2)

Data Set 1

Interstitial pattern

0.92 (0.9–0.98)

0.95 (0.52–1)

0.77 (0.71–0.83)

3.88 (2.8–5.5)

0.14 (0.02–0.9)

Data Set 2

Interstitial pattern

0.79 (0.73–0.85)

0.95 (0.87–1)

0.44 (0.4–0.48)

1.72 (1.6–1.9)

0.09 (0.02–0.3)

Data Set 1

Mass

0.77 (0.68–0.875)

0.95 (0.74–1)

0.42 (0.35–0.5)

1.64 (1.4–1.9)

0.12 (0.02–0.8)

Data Set 2

Mass

0.66 (0.55–0.77)

0.95 ( 0.85–1)

0.11 (0.09–0.14)

1.1 (1–1.2)

0.26 (0.04–1.8)

Data Set 1

Megaesophagus

0.78 (0.56–1)

0.95 (0.42–1)

0.29 (0.17–0.27)

1.10 (0.8–1.5)

0.65(0.1–4.1)

Data Set 2

Megaesophagus

0.80 (0.71–0.90)

0.95 (0.76–1)

0.31 (0.27–0.34)

1.37 (1.2–1.5)

0.16 (0.02–1.1)

Data Set 1

Pleural effusion

0.96 (0.9–1)

0.95 (0.64–1)

0.57 (0.49–0.63)

2.11 (1.7–2.6)

0.14 (0.02–0.9)

Data Set 2

Pleural effusion

0.96 (0.93–0.98)

0.95 (0.73–1)

0.81 (0.77–0.84)

4.87(4.0–5.9)

0.07 (0.01–0.5)

Data Set 1

Pneumothorax

0.88 (0.72–0.96)

0.95 (0.75–0.98)

0.40 (0.35–0.34)

1.56 (1.3–1.6)

0.24 (0.07–1.8)

Data Set 2

Pneumothorax

0.84 (0.72–0.96)

0.95 (0.64–0.96)

0.30 (0.27–0.34)

1.35 (1.2–1.5)

0.18 (0.03–1.2)

Data Set 1

Unremarkable

0.88 (0.83–0.92)

0.95 (0.89–0.98)

0.63 (0.54–0.73)

2.62 (2–4.4)

0.08 (0.04–0.2)

Data Set 2

Unremarkable

0.83 (0.80–0.86)

0.95 (0.92–0.97)

0.44 (0.38–0.5)

1.69 (1.5–1.9)

0.11(0.07–0.2)

  1. AUC area under the receiver operator curve, PLR positive likelihood ratio, NLR negative likelihood ratio.
  2. Most relevant results have been bolded.