TABLE 2.
All Reference Standard Positive, n = 50 | All Reference Standard Negative, n = 138 | ||||||
---|---|---|---|---|---|---|---|
n | TP | FN | TN | FP | Specificity % | Sensitivity % | |
Experienced readers | 18 | ||||||
Reads without ML support | 188 | 35 | 15 | 121 | 17 | 87.7 | 70.0 |
Reads with ML support | 188 | 33 | 17 | 119 | 19 | 86.2 | 66.0 |
Difference in proportions | −1.5 (95% CI, −6.4, 3.5; P = 0.387) | −4.0 (95% CI, −13.5, 5.5; P = 0.344) | |||||
Subset Reference Standard Positive, n = 15 | Subset Reference Standard Negative, n = 38 | ||||||
Inexperienced readers | 7 | ||||||
Reads without ML support | 53 | 9 | 6 | 29 | 9 | 76.3 | 60.0 |
Reads with ML support | 53 | 11 | 4 | 29 | 9 | 76.3 | 73.3 |
Difference in proportions | 0.0 (95% CI, −15.0, 15.0; P = 0.613) | 13.3 (95% CI, −7.9, 34.5; P = 0.313) |
Per-patient sensitivity and specificity for experienced and inexperienced WB-MRI readers.
ML, machine learning.