Table 2 Comparison of image quantification accuracies of 92 whole lung HRCT images between the CNNs with and without Perlin noise and the radiologists.
From: A Perlin Noise-Based Augmentation Strategy for Deep Learning with Small Data Samples of HRCT Images
DILD class | Inter-radiologist agreement(%) | Radiologist 1 | Radiologist 2 | ||||
|---|---|---|---|---|---|---|---|
CNN-P (%) | CNN-C (%) | P-value | CNN-P (%) | CNN-C (%) | P-value | ||
Normal | 62.7–92.2 | 40.2 | 36.1 | <0.001 | 47.6 | 42.9 | <0.001 |
GGO | 15.7–67.5 | 38.9 | 33.8 | <0.001 | 29.4 | 23.0 | <0.001 |
Consolidation | 52.3–60.4 | 48.6 | 38.5 | <0.001 | 59.1 | 48.2 | <0.001 |
RO | 45.7–56.0 | 67.3 | 62.2 | <0.001 | 67.8 | 61.5 | <0.001 |
Emphysema | 18.2–70.0 | 71.0 | 57.2 | 0.005 | 37.2 | 31.0 | 0.69 |
Honeycombing | 39.5–82.2 | 63.7 | 71.6 | <0.001 | 57.7 | 67.6 | <0.001 |
Mean | 45.3–65.1 | 55.0 | 49.9 | <0.001 | 49.8 | 45.7 | <0.001 |
