Table 1 Comparison between manual method and three segmentation approaches for quantitation of cerebral microhemorrhages.
From: Spectroscopic and deep learning-based approaches to identify and quantify cerebral microhemorrhages
Analysis process | Average time commitment per image | Modifiable | Sensitivity, specificity | Intraclass correlation coefficient (95% confidence interval) | Absolute area difference (µm2) (interquartile range) | Percent area difference (%) (interquartile range) | |
|---|---|---|---|---|---|---|---|
Manual approach | Subjective | 10 min | N/A | N/A | N/A | N/A | N/A |
Ratiometric approach | Semi-automated | 15 s | Easy | 0.835, 0.997 | 0.992 (0.989–0.995) | 75.2 (33.3–172.3) | 11.0 (4.7–21.2) |
Phasor approach | Semi-automated | 30 s | Moderate | 0.768, 0.998 | 0.993 (0.977–0.997) | 124.5 (55.2–255.2) | 18.8 (9.3–29.4) |
Deep learning approach | Automated | 3 s | Difficult | 0.708, 0.998 | 0.961 (0.915–0.979) | 71.1 (33.6–167.7) | 12.7 (8.8–18.9) |
