Table 4 The performances of 12 pathological feature detection models
From: Oral epithelial dysplasia detection and grading in oral leukoplakia using deep learning
Pathological Feature | Accuracy (%) (95% CI) | AUC (95% CI) |
|---|---|---|
Irregular epithelial stratification | 78.2(67.3 ∼ 89.1) | 0.768 (0.576 ∼ 0.930) |
Loss of polarity of basal cells | 79.3 (62.1 ∼ 93.1) | 0.722 (0.578 ∼ 0.813) |
Drop-shaped rete ridges | 86.9 (78.7 ∼ 95.1) | 0.900 (0.814 ∼ 0.967) |
Increased number of mitotic figures | 81.7 (71.7 ∼ 91.7) | 0.806 (0.671 ∼ 0.927) |
Premature keratinization in single cells | 96.9 (92.2 ∼ 98.4) | 0.734 (0.544 ∼ 0.856) |
Loss of epithelial cell cohesion | 98.4 (93.4 ∼ 98.4) | 0.983 (0.948 ∼ 1.000) |
Abnormal variation in nuclear size | 67.6 (51.4 ∼ 81.1) | 0.725 (0.612 ∼ 0.818) |
Abnormal variation in nuclear shape | 78.9 (65.8 ∼ 92.1) | 0.776 (0.600 ∼ 0.923) |
Abnormal variation in cell size | 80.0 (68.9 ∼ 91.1) | 0.883 (0.737 ∼ 0.993) |
Abnormal variation in cell shape | 88.6 (79.5 ∼ 97.7) | 0.764 (0.646 ∼ 0.876) |
Increased N: C ratio | 62.5 (51.5 ∼ 79.5) | 0.742 (0.604 ∼ 0.796) |
Hyperchromasia | 65.4 (51.9 ∼ 78.8) | 0.679 (0.587 ∼ 0.759) |