Table 3.
Summary Estimates of DL performance in retinal vessel segmentation.
| SE with 95% CI | SP with 95% CI | LR+ with 95% CI | LR− with 95% CI | DOR with 95% CI | |
|---|---|---|---|---|---|
| DRIVE | |||||
| Human experts | 0.77 | 0.97 | NR | NR | NR |
| DL * | 0.77 (0.77–0.77) | 0.97 (0.97–0.97) | 28.19 (24.21–32.82) | 0.23 (0.22–0.25) | 120.57 (99.66–145.86) |
| STARE | |||||
| Human experts | 0.89 | 0.93 | NR | NR | NR |
| DL * | 0.79 (0.79–0.79) | 0.97 (0.97–0.97) | 31.02 (30.77–31.28) | 0.21 (0.21–0.21) | 136.67 (135.42–137.0) |
| CHASE_DB1 | |||||
| Human experts | 0.83 | 0.97 | NR | NR | NR |
| DL * | 0.78 (0.78–0.78) | 0.97 (0.97–0.97) | 22.97 (22.75–23.20) | 0.23 (0.23–0.23) | 109.27 (108.0–110.56) |
| HRF | |||||
| Human experts | NR | NR | NR | NR | NR |
| DL * | 0.81 (0.81–0.81) | 0.92 (0.92–0.92) | 10.32 (10.26–10.38) | 0.21 (0.21–0.21) | 51.75 (51.35–52.16) |
* Note: DL = Deep Learning, NR = Not Reported, SE = Sensitivity, SP = Specificity, LR = Likelihood Ratio, CI = Confidence Interval, * = Summarized.