Machine learning-enabled prediction of hemorrhagic transformation post-thrombectomy using quantitative DSA.

Area of Science:

• Medical Imaging
• Neurology
• Machine Learning

Background:

• Hemorrhagic transformation (HT) is a significant complication following mechanical thrombectomy (MT) for acute ischemic stroke.
• Predicting HT is crucial for optimizing treatment strategies and improving patient outcomes.
• Quantitative digital subtraction angiography (qDSA) offers detailed hemodynamic insights.

Purpose of the Study:

• To develop and validate a predictive model for post-thrombectomy HT.
• To integrate hemodynamic features from qDSA with clinical data using machine learning.
• To assess the predictive performance of machine learning models for HT.

Main Methods:

• Retrospective analysis of 171 patients with acute anterior circulation large-vessel occlusion undergoing MT.
• Extraction of 39 hemodynamic parameters from postoperative qDSA perfusion images.
• Application of 5 feature selection algorithms and 5 machine learning models (including Elastic-Logistic).
• Evaluation using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC).
• Model interpretation using SHapley Additive exPlanations (SHAP).

Main Results:

• The best-performing model, integrating qDSA and clinical features, achieved an average AUC of 0.86.
• A model using qDSA features alone reached an average AUC of 0.81.
• Hemorrhagic transformation occurred in 68 out of 171 patients.

Conclusions:

• Machine learning models integrating qDSA-derived hemodynamic and clinical features can effectively predict post-thrombectomy HT.
• This approach offers a preliminary tool for predicting HT in acute ischemic stroke patients undergoing MT.
• Further validation is warranted to refine the predictive capabilities.