Table 2 Performance of our proposed liver trauma segmentation approach stratified based on the severity of the injury as well as the performance of the baseline U-net architecture. Numbers in parentheses are standard deviations
From: A deep learning framework for automated detection and quantitative assessment of liver trauma
Method | % Liver disruption | Dice (%) | Recall (%) | Precision (%) | RVD (%) | VOE (%) |
|---|---|---|---|---|---|---|
Proposed method | 0–2% (n = 15) | 28.06 (23.09) | 32.15 (32.90) | 35.66 (27.01) | 15.83 (113.41) | 81.68 (16.01) |
Proposed method | 2–5% (n = 4) | 58.35 (17.91) | 54.78 (25.39) | 66.21 (6.93) | − 19.16 (30.85) | 57.05 (18.64) |
Proposed method |  > 5% (n = 15) | 72.45 (11.82) | 73.84 (18.60) | 75.35 (11.51) | 1.86 (36.03) | 42.02 (13.64) |
Proposed method | All (n = 34) | 51.21 (27.74) | 53.20 (32.56) | 56.76 (27.21) | 5.55 (78.88) | 61.29 (24.07) |
U-net (no post-processing) | All (n = 34) | 47.75 (27.61) | 47.32 (31.13) | 56.64 (28.97) | − 2.71 (63.10) | 64.50 (23.80) |