Advanced feature fusion of radiomics and deep learning for accurate detection of wrist fractures on X-ray images DOI Creative Commons

Mohamed J. Saadh,

Qusay Mohammed Hussain,

Rafid Jihad Albadr

и другие.

BMC Musculoskeletal Disorders, Год журнала: 2025, Номер 26(1)

Опубликована: Май 20, 2025

The aim of this study was to develop a hybrid diagnostic framework integrating radiomic and deep features for accurate reproducible detection classification wrist fractures using X-ray images. A total 3,537 images, including 1,871 fracture 1,666 non-fracture cases, were collected from three healthcare centers. Radiomic extracted the PyRadiomics library, derived bottleneck layer an autoencoder. Both feature modalities underwent reliability assessment via Intraclass Correlation Coefficient (ICC) cosine similarity. Feature selection methods, ANOVA, Mutual Information (MI), Principal Component Analysis (PCA), Recursive Elimination (RFE), applied optimize set. Classifiers such as XGBoost, CatBoost, Random Forest, Voting Classifier used evaluate performance. dataset divided into training (70%) testing (30%) sets, metrics accuracy, sensitivity, AUC-ROC evaluation. combined approach consistently outperformed standalone methods. paired with MI achieved highest performance, test accuracy 95%, sensitivity 94%, 96%. end-to-end model competitive results 93% 94%. SHAP analysis t-SNE visualizations confirmed interpretability robustness selected features. This demonstrates potential enhance performance forearm fractures, providing reliable interpretable solution suitable clinical applications.

Язык: Английский

Advanced feature fusion of radiomics and deep learning for accurate detection of wrist fractures on X-ray images DOI Creative Commons

Mohamed J. Saadh,

Qusay Mohammed Hussain,

Rafid Jihad Albadr

и другие.

BMC Musculoskeletal Disorders, Год журнала: 2025, Номер 26(1)

Опубликована: Май 20, 2025

The aim of this study was to develop a hybrid diagnostic framework integrating radiomic and deep features for accurate reproducible detection classification wrist fractures using X-ray images. A total 3,537 images, including 1,871 fracture 1,666 non-fracture cases, were collected from three healthcare centers. Radiomic extracted the PyRadiomics library, derived bottleneck layer an autoencoder. Both feature modalities underwent reliability assessment via Intraclass Correlation Coefficient (ICC) cosine similarity. Feature selection methods, ANOVA, Mutual Information (MI), Principal Component Analysis (PCA), Recursive Elimination (RFE), applied optimize set. Classifiers such as XGBoost, CatBoost, Random Forest, Voting Classifier used evaluate performance. dataset divided into training (70%) testing (30%) sets, metrics accuracy, sensitivity, AUC-ROC evaluation. combined approach consistently outperformed standalone methods. paired with MI achieved highest performance, test accuracy 95%, sensitivity 94%, 96%. end-to-end model competitive results 93% 94%. SHAP analysis t-SNE visualizations confirmed interpretability robustness selected features. This demonstrates potential enhance performance forearm fractures, providing reliable interpretable solution suitable clinical applications.

Язык: Английский

Процитировано

0