Machine learning-based Cerebral Venous Thrombosis diagnosis with clinical data.

Area of Science:

• Neurology
• Medical Informatics
• Machine Learning

Background:

• Cerebral Venous Thrombosis (CVT) presents diagnostic challenges due to varied symptoms and disease progression.
• Early diagnosis is crucial for effective CVT prognosis.
• A need exists for efficient screening tools to aid in CVT diagnosis.

Purpose of the Study:

• To develop and validate a machine learning-based screening algorithm for Cerebral Venous Thrombosis (CVT).
• To utilize clinical and demographic data for CVT diagnosis.
• To assess the performance of different machine learning models in CVT screening.

Main Methods:

• Utilized data from the Iran Cerebral Venous Thrombosis Registry (ICVTR) including 314 CVT cases and 575 controls.
• Collected and analyzed 60 clinical and demographic features.
• Evaluated generalized linear model, random forest, support vector machine, and extreme gradient boosting algorithms, incorporating missing value imputation.

Main Results:

• The support vector machine model achieved the highest Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.910 when all variables with imputed missing values were used.
• The support vector machine model demonstrated the best recall (0.77) when using variables with less than 50% missing data.
• The random forest model achieved the highest precision (0.94) using variables with less than 50% missing data.

Conclusions:

• Machine learning techniques applied to clinical data show significant promise for accurate CVT diagnosis.
• This ML-based approach can serve as a valuable assistive tool in clinical settings.
• The developed algorithm may offer an alternative to resource-intensive imaging techniques for CVT screening.