Table 2. Results of the four readings and experts' agreement of the two 228-frame datasets.
Dataset | Reader | Reading | Adequately cleansed frames (%) |
Inadequately cleansed frames (%) |
Pearson’s coefficient | Exclusion of images for inter-reader discrepancies 1 |
1 | 1 | 1 | 134 (46.5 %) | 154 (53.5 %) | Intra-reader 1 = 0.90 Intra-reader 2 = 0.85 Inter-reader 1 – 2 = 0.87 |
8 frames |
2 | 126 (43.7 %) | 162 (56.3 %) | ||||
2 | 1 | 121 (42.0 %) | 167 (58.0 %) | |||
2 | 126 (43.7 %) | 162 (56.3 %) | ||||
2 | 3 | 1 | 121 (42.0 %) | 167 (58.0 %) | Intra-reader 3 = 0.89 Intra-reader 4 = 0.81 Inter-reader 3 – 4 = 0.82 |
11 frames |
2 | 119 (41.3 %) | 169 (58.7 %) | ||||
4 | 1 | 104 (36.1 %) | 184 (63.9 %) | |||
2 | 110 (38.2 %) | 178 (61.8 %) |
Three or four agreements among the four readings of the same frame made a definitive classification. For any discrepancy (2 adequate and 2 inadequate classifications of the same frame), image was excluded from the analysis.