Machine learning in developing a predictive model for chronic hydrocephalus following aneurysmal subarachnoid...

Area of Science:

• Neurosurgery
• Radiology
• Artificial Intelligence

Background:

• Aneurysmal subarachnoid hemorrhage (aSAH) poses a significant risk for developing chronic hydrocephalus.
• Accurate prediction of chronic hydrocephalus is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate a predictive model for chronic hydrocephalus in aSAH patients.
• To integrate clinical data, radiomics features, and hematoma volume using machine learning for enhanced prediction accuracy.

Main Methods:

• Utilized a dataset of 410 aSAH patients, collecting clinical and imaging data.
• Developed a 3D-Unet model for accurate hematoma volume quantification.
• Extracted radiomic features and employed LASSO regression for feature selection, constructing a clinical-radiological nomogram using machine learning algorithms.

Main Results:

• The 3D-Unet model achieved high accuracy in SAH volume segmentation (DSC: 0.85).
• Identified six independent predictive factors, including hematoma volume and periventricular white matter changes.
• The logistic regression-based nomogram demonstrated strong predictive performance (AUC: 0.884 in training, 0.860 in test set) with good calibration.

Conclusions:

• The developed clinical-radiological nomogram effectively predicts the risk of chronic hydrocephalus in aSAH patients.
• The integration of hematoma volume, clinical, and radiomic features significantly improves prediction accuracy.
• The model demonstrates clinical utility and potential for guiding patient management.