Table 7.
Inter rater agreement for the MLO projections for both readers and the dCNN compared to each other and to the consensus decision
| RMLO Projections | ACR A | ACR B | ACR C | ACR D | Overall | |||||
| κ (95% CI) |
Agreement | κ (95% CI) |
Agreement | κ (95% CI) |
Agreement | κ (95% CI) |
Agreement | κ (95% CI) |
Agreement | |
| Reader 1/Reader 2 | 0.57 (0.37–0.77) |
Moderate | 0.69 (0.55–0.85) |
Strong | 0.40 (0.22–0.60) |
Fair | 0.66 (0.48–0.85) |
Strong | 0.67 (0.58–0.76) |
Strong |
| Reader 1/Consensus | 0.81 (0.66–0.97) |
Almost perfect | 0.79 (0.66–0.92) |
Almost perfect | 0.75 (0.61–0.91) |
Strong | 0.91 (0.82–1.00) |
Almost perfect | 0.87 (0.81–0.93) |
Almost perfect |
| Reader 2/Consensus | 0.76 (0.60–0.91) |
Strong | 0.90 (0.80–0.99) |
Almost perfect | 0.61 (0.47–0.80) |
Strong | 0.75 (0.60–0.92) |
Strong | 0.80 (0.73–0.88) |
Strong |
| dCNN/Consensus | 0.84 (0.70–0.97) |
Almost perfect | 0.81 (0.69–0.93) |
Strong | 0.87 (0.77–0.98) |
Almost perfect | 0.94 (0.86–1.00) |
Almost perfect | 0.91 (0.86–0.96) |
Almost perfect |
ACR, American College of Radiology; dCNN, deep convolutional neural network; MLO, medio-lateral oblique;