A classification model for concurrent posterior circulation infarction in vertebrobasilar dolichoectasia using...

Area of Science:

• Neurology
• Medical Imaging
• Machine Learning

Background:

• Vertebrobasilar dolichoectasia (VBD) is a cerebrovascular condition often associated with posterior circulation infarction (PCI).
• Lack of tools to classify concurrent PCI in VBD patients necessitates novel approaches.
• This study focuses on developing a machine learning model for this classification task.

Purpose of the Study:

• To develop and validate a machine learning-based classification model.
• To identify concurrent posterior circulation infarction (PCI) in patients with vertebrobasilar dolichoectasia (VBD).
• To provide a tool for clinical decision-making in VBD patients.

Main Methods:

• Retrospective data collection from 380 participants.
• Random forest and XGBoost models were trained and validated.
• Boruta feature selection and SHAP analysis were employed to identify key predictors.

Main Results:

• The XGBoost model achieved an AUC of 0.884, demonstrating strong performance.
• Key predictors for concurrent PCI included basilar artery diameter (BAD) and midline distance (BAMD).
• The model integrated clinical and imaging variables for classification.

Conclusions:

• A machine learning model was developed and internally validated to classify concurrent PCI in VBD.
• External validation and prospective studies are crucial for clinical implementation.
• A web calculator is available to aid in identifying individuals with concurrent PCI.