Table 1 Summary of the test datasets.

From: The degradation of performance of a state-of-the-art skin image classifier when applied to patient-driven internet search

 

RD

SNU

Edinburgh

TeleDerm

Number of cases

1282

2101

1300

340

Source

Internet community

Tertiary care center

Tertiary care center

Teledermatology

Photographer

Patient

Physician

Professional photographer

Patient

Fitzpatrick skin type

3–4

1–2

1–4

Number of disease classes

62

133

10

87

Disease category

Inflammatory Dermatitis

12 (0.9%)

131 (6.2%)

82 (24.1%)

Acne/rosacea

51 (2.4%)

78 (22.9%)

Autoimmune

2 (0.2%)

90 (4.3%)

34 (10.0%)

Papulosquamous

105 (5.0%)

17 (5.0%)

Others inflammatory

30 (2.3%)

159 (7.6%)

14 (4.1%)

Viral infection

39 (3.0%)

144 (6.9%)

22 (6.5%)

Fungal infection

7 (0.5%)

85 (4.0%)

20 (5.9%)

Bacterial infection

3 (0.2%)

125 (5.9%)

9 (2.6%)

Parasitic infection

15 (0.7%)

1 (0.3%)

Benign neoplastic

896 (69.9%)a

620 (29.5%)

819 (63.0%)

28 (8.2%)

Malignant neoplastic

123 (9.6%)a

182 (8.7%)

481 (37.0%)

4 (1.2%)

Alopecia, scarring

8 (2.4%)

Alopecia, non-scarring

20 (1.0%)

7 (2.1%)

Others

170 (13.3%)

374 (17.8%)

16 (4.7%)

  1. aThe ground truth of the RD dataset was voted on by five specialists, whereas the malignancies in the other datasets were determined by pathological examinations.