Table 2.
Diagnostic performance for sarcopenia prediction in the development set (%, μ ± σ).
| Models | ACC | SEN | SPE | PPV | NPV | AUC | P-value* | P-value** |
|---|---|---|---|---|---|---|---|---|
| Clinical | 70.57 ± 5.00 | 79.09 ± 6.14 | 68.98 ± 6.83 | 32.83 ± 4.20 | 94.72 ± 0.98 | 74.53 ± 2.70 | ||
| USG | 69.00 ± 4.52 | 78.18 ± 4.69 | 67.29 ± 5.98 | 31.26 ± 3.72 | 94.34 ± 0.87 | 69.88 ± 2.31 | ||
| FLF | 70.29 ± 6.63 | 72.73 ± 9.58 | 69.83 ± 9.35 | 31.74 ± 3.78 | 93.40 ± 1.66 | 68.78 ± 4.34 | 0.836 | 0.665 |
| SLF-SR | 76.71 ± 3.81 | 73.64 ± 2.87 | 77.29 ± 4.94 | 38.18 ± 4.02 | 94.03 ± 0.44 | 75.96 ± 1.89 | < 0.001+ | 0.006+ |
| SLF-LSE | 66.00 ± 6.42 | 85.45 ± 7.67 | 62.37 ± 8.55 | 30.35 ± 3.67 | 96.01 ± 1.49 | 74.16 ± 2.34 | 0.006+ | 0.224 |
| SLF-AAC | 67.29 ± 6.82 | 82.73 ± 10.9 | 64.41 ± 9.69 | 30.91 ± 3.85 | 95.51 ± 2.18 | 74.58 ± 2.66 | 0.059 | 0.525 |
| SLF-RF | 67.57 ± 6.39 | 76.36 ± 9.77 | 65.93 ± 8.13 | 30.19 ± 5.85 | 93.81 ± 2.13 | 69.57 ± 5.87 | 0.128 | 0.607 |
| SLF-SVM | 71.14 ± 13.6 | 46.36 ± 24.8 | 75.76 ± 20.1 | 32.23 ± 10.8 | 88.95 ± 2.93 | 48.69 ± 12.6 | 0.913 | 0.701 |
| SLF-AB | 73.43 ± 6.57 | 65.45 ± 16.5 | 74.92 ± 10.3 | 34.22 ± 6.04 | 92.41 ± 2.65 | 61.36 ± 6.96 | 0.413 | 0.046+ |
*μ = mean, σ = standard deviation, ACC = accuracy, SEN = sensitivity, SPE = specificity, PPV = positive predictive value, NPV = negative predictive value, and AUC = area under the receiver operating characteristics (ROC) curve, FLF = feature-level fusion, SLF-SR = score-level fusion by the sum-rule, SLF-LSE = score-level fusion by the least squares estimation, SLF-AAC = score-level fusion by area above the ROC curve optimization, SLF-RF = score-level fusion by random forest, SLF-SVM = score-level fusion by support vector machine, SLF-AB = score-level fusion by AdaBoost.
* Comparison of accuracy between the clinical features and after fusion (paired t-test).
** Comparison of accuracy between the USG features and after fusion (paired t-test).
+ Statistically significant.