Machine learning based classification of spontaneous intracranial hemorrhages using radiomics features

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Spontaneous intracerebral hemorrhage (ICH) has multiple etiologies, including primary, tumoral, and vascular malformations.
• Accurate differentiation of ICH causes is crucial for appropriate patient management and treatment strategies.
• Non-contrast computed tomography (NCCT) is a primary imaging modality for ICH detection.

Purpose of the Study:

• To evaluate the effectiveness of radiomics features from NCCT scans in distinguishing between different causes of spontaneous ICH.
• To develop and validate a machine learning model for classifying ICH etiologies based on imaging biomarkers.

Main Methods:

• A cohort of 141 ICH patients with primary, tumoral, or vascular malformation etiologies was analyzed.
• Radiomics features were extracted from initial NCCT scans, and mutual information was used for feature selection.
• An AdaBoost hierarchical classification model was employed, incorporating patient age and ICH location.

Main Results:

• The classification model achieved an overall accuracy of 0.79.
• High sensitivity and specificity were reported for differentiating primary (0.86/0.87), tumoral (0.78/0.93), and vascular malformation (0.72/0.89) related ICH.
• Area under the curve (AUC) values ranged from 0.82 to 0.86, with texture-based features proving important.

Conclusions:

• Machine learning models utilizing radiomics features show promise for classifying non-traumatic ICH etiologies.
• This approach can potentially assist radiologists in selecting appropriate diagnostic workup pathways.
• Radiomics analysis of NCCT may improve diagnostic accuracy and guide clinical decision-making for ICH patients.