Table 1.
Classification of pelvic ring fractures according to AO/OTA classification system.
Table 2.
The number of X-ray data and normal data collected by pelvic fracture type.
Fig 1.
(a) a collection pelvic X-ray image, (b) a pelvic region ROI image.
Fig 2.
(a) in case the pelvic area is not clearly visible due to severe blur and (b) in case the pelvic area is not clearly visible due to organs and gases.
Table 3.
Datasets used for training and testing of machine learning models.
Fig 3.
Image and histogram changes before and after application of the histogram equalization algorithm.
Fig 4.
Radiomics feature extraction is performed for the pelvic ring region.
18 first-order features, 75 second-order features. Second-order features included 24 GLCM, 16 GLRLM, 16 GLSZM, 5 NGTDM, and 14 GLDM.
Fig 5.
After radiomics feature extraction, the feature selection method is used to select meaningful features for training the machine learning model.
Learning proceeds based on selected features. Comparison of specialist’s reading findings and machine learning model prediction results.
Fig 6.
AUC for each combination of LR, SVM, RF, XGB, MLP, KNN, and LGBM machine learning models and RFE, SFS, LASSO, and Ridge feature selection methods through heatmap.
The higher the heatmap value, the closer it is to black, and the higher the performance for the classification of instability of pelvic fracture. LR, logistic regression; SVM, support vector machine; RF, random forest; XGB, extreme gradient boosting; MLP, multi-layer perceptron; KNN, k-nearest neighbor; LGBM, light gradient boosting machine; RFE, recursive feature elimination; SFS, sequential feature selection.
Fig 7.
ROC analysis results of machine learning models LR, SVM, RF, XGB, MLP, KNN, and LGBM for classification of normal and pelvic fracture types when using feature selection method RFE.
Fig 8.
A graph of the average feature importance of feature selection method RFE and each machine learning model combination.
The higher the heatmap value, the closer it is to black, and the higher the impact on the instability classification of pelvic fracture.
Fig 9.
Feature importance graph for feature selection method RFE and each machine learning model: LR, SVM, RF, MLP, XGB, KNN, and LGBM.
The higher the heatmap value, the closer it is to black, and the higher the impact on the instability classification of pelvic fracture.