Table 3 Comparing the performance of the classification in identifying soft tissue tumor differentiation in previous and current studies
Study | Number of patients with lipomas in benign group | Number of patients with ALTs/WDLSs in malignant group | Classify | Machine learning algorithm | AUC | Sensitivity (%) | Specificity (%) | Accuracy (%) |
|---|---|---|---|---|---|---|---|---|
Cay, 2022 [15] | 45 | 20/20 (100%) | Lipomas vs. ALTs/WDLSs | SVM | 0.98 | 96.8 | 93.7 | – |
Fradet, 2022 [16] | 40 | 45/45 (100%) | Lipomas vs. ALTs | Logistic regression | 0.50 | 100 | 0.0 | – |
|  |  |  |  | SVM | 0.47 | 70.0 | 32.0 | – |
|  |  |  |  | Random forest | 0.71 | 64.0 | 68.0 | – |
|  |  |  |  | Gradient boosting | 0.70 | 67.0 | 64.0 | – |
Tang, 2022 [20] | 90 | 32/32 (100%) | Lipomas vs. ALTs | Random forest | 0.94 | 85.7 | 100 | 96.0 |
Yang, 2022 [23] | 69 | 58/58 (100%) | Lipomas vs. WDLSs | SVM | 0.95 | 95.0 | 88.9 | 92.1 |
Malinauskaite, 2020 [18] | 24 | 5/14 (35.7%) | Lipomas vs. Liposarcomas | SVM | 0.93 | 88.8 | 100 | 94.7 |
|  |  |  |  | LDA | 0.93 | – | – | 89.5 |
|  |  |  |  | Naïve Bayes | 0.81 | – | – | 79.0 |
|  |  |  |  | Logistic regression | 0.812 | – | – | 73.7 |
Leporq, 2020 [17] | 40 | ?/41 | Lipomas vs. Liposarcomas | SVM | 0.96 | 100 | 90 | 95.0 |
Vos, 2019 [22] | 58 | 58/80 (71.6%) | Lipomas vs. WDLSs | Various* | 0.83 | 68 | 84 | 67.0 |
Thornhill, 2014 [21] | 24 | ?/20 | Lipomas vs. Liposarcomas | LDA | N/A | 85 | 96 | 91.0 |
Current study | ||||||||
 Machine learning model | 38 | 30/30 (100%) | IM lipomas vs. ALTs/WDLSs | LASSO LR | 0.88 | 91.6 | 85.7 | 89.0 |
 MSK radiologist 1 | 38 | 30/30 (100%) | IM lipomas vs. ALTs/WDLSs | - | 0.94 | 97.4 | 90.9 | 95.0 |
 MSK radiologist 2 | 38 | 30/30 (100%) | IM lipomas vs. ALTs/WDLSs | - | 0.91 | 100 | 81.8 | 93.3 |