Prediction model for deep vein thrombosis stability based on multiple machine learning methods.

Area of Science:

• Medical Imaging
• Machine Learning
• Vascular Medicine

Background:

• Deep vein thrombosis (DVT) poses a significant risk, with stability being a crucial factor in predicting outcomes like acute pulmonary embolism (APE).
• Accurate prediction of DVT stability is essential for timely and appropriate clinical intervention.

Purpose of the Study:

• To develop and compare multiple machine learning (ML) models for predicting DVT stability.
• To identify the optimal model utilizing clinical and computed tomography (CT) texture features for DVT stability prediction.

Main Methods:

• 108 patients with DVT were analyzed, categorizing them into stable (no APE) and unstable (with APE) groups.
• CT texture features were extracted from thrombus regions, and clinical data were collected.
• Four ML algorithms (logistic regression, SVM, KNN, XGBoost) were trained and validated using a combination of clinical and CT texture features.

Main Results:

• The logistic regression (LR) model achieved the highest prediction performance.
• LR model metrics included an AUC of 0.87, accuracy of 0.79, precision of 0.75, recall of 0.80, and specificity of 0.87.
• The LR model demonstrated superior predictive capability compared to SVM, KNN, and XGBoost models.

Conclusions:

• Machine learning models integrating clinical and CT texture features can reliably predict DVT stability.
• The developed ML models, particularly the LR model, offer a promising tool for assessing DVT stability and guiding patient management.