Table 3.
Accuracy of the AI algorithm, dermatologists, dermatology residents, and general practitioners of diseases included in the algorithm (‘in-distribution’) training dataset
| Category | n | Top-1 Accuracy | Top-3 Accuracy | |||
|---|---|---|---|---|---|---|
| AI algorithm | Average dermatologist | Average resident | Average general practitioner | AI algorithm | ||
| General | 305 | 45.3% | 63.1% | 60.9% | 52.9% | 69.2% |
| Balanced | 305 | 47.6% | 49.7% | 47.7% | 39.7% | 72.2% |
| Inflammatory | 210 | 40.5% | 65.7% | 61.3% | 53.0% | 67.1% |
| Dermatitis | 80 | 35.0% | 62.5% | 52.9% | 50.0% | 60% |
| Acne/rosacea | 75 | 49.3% | 81.3% | 81.8% | 74.6% | 80% |
| Autoimmune | 30 | 30.0% | 42.2% | 41.1% | 18.9% | 46.7% |
| Papulosquamous | 17 | 52.9% | 52.9% | 47.1% | 33.3% | 82.4% |
| Infectious | 47 | 55.3% | 58.2% | 59.6% | 56.7% | 72.3% |
| Viral | 22 | 50.0% | 65.1% | 62.1% | 51.5% | 68.2% |
| Fungal | 18 | 61.1% | 64.8% | 70.4% | 79.6% | 77.8% |
| Bacterial | 7 | 57.1% | 19.1% | 23.8% | 14.3% | 71.4% |
| Neoplastic | 28 | 39.3% | 52.4% | 54.7% | 44.1% | 64.3% |
| Benign | 24 | 45.8% | 59.7% | 62.5% | 48.6% | 70.8% |
| Malignant | 4 | 0% | 8.3% | 8.3% | 16.7% | 25% |
| Alopecia | 5 | 100% | 80% | 80% | 66.7% | 100% |
| Scarring | 0 | N/A | N/A | N/A | N/A | N/A |
| Non-scarring | 5 | 100% | 80% | 80% | 66.7% | 100% |
| Other | 15 | 73.3% | 55.5% | 64.4% | 51.1% | 86.7% |
AI, Artificial Intelligence; D, dermatologist; R, residents; GP, general practitioners.