Pelvic Injury Discriminative Model Based on Data Mining Algorithm.

Area of Science:

• Medical imaging analysis
• Machine learning in healthcare
• Radiology and diagnostics

Background:

• Pelvic CT imaging is crucial for diagnosing injuries.
• Effective feature extraction and dimension reduction are vital for accurate analysis.
• Current methods may benefit from advanced computational approaches.

Purpose of the Study:

• To reduce feature dimension from pelvic CT images using Principal Component Analysis (PCA) and Partial Least Squares (PLS).
• To develop a Support Vector Machine (SVM) model for pelvic injury identification using reduced dimension data.
• To evaluate the feasibility of this data mining model for clinical application.

Main Methods:

• A dataset of 146 pelvic CT images (normal and injured) was used.
• 80% of images served as the training set, 20% as the testing set.
• Methods included image preprocessing, feature extraction, dimension reduction (PCA/PLS), and SVM model establishment.

Main Results:

• Partial Least Squares (PLS) outperformed Principal Component Analysis (PCA) for dimension reduction.
• Support Vector Machine (SVM) demonstrated superior performance over Naive Bayesian Classifier (NBC).
• The SVM model, using 12 PLS factors, achieved 100% accuracy on the training and cross-validation sets, and 93.33% on the testing set.

Conclusions:

• A data mining model based on CT images can accurately identify pelvic injuries.
• The developed model demonstrates high accuracy, supporting automated and rapid pelvic injury detection.
• This approach provides a foundation for advanced diagnostic tools in radiology.