SWI and CTP fusion model based on sparse representation method to predict cerebral infarction trend

Area of Science:

• Radiology
• Medical Imaging
• Neurology

Background:

• Susceptibility Weighted Imaging (SWI) signals are linked to venous reflux disorder and perfusion defects.
• Computed Tomography Perfusion (CTP) provides spatiotemporal perfusion data.
• SWI and CTP offer complementary information crucial for predicting cerebral infarction outcomes.

Purpose of the Study:

• To evaluate the combined predictive power of SWI and CTP for cerebral infarction prognosis.
• To assess the accuracy of a machine learning model integrating SWI and CTP features.
• To compare the predictive performance of integrated SWI+CTP versus individual modalities.

Main Methods:

• A retrospective study extracted SWI and CTP radiomic features from 66 patients.
• A machine learning model using sparse representation was developed and tested on 33 prospective patients.
• Model predictions were validated against Diffusion Weighted Imaging (DWI) follow-up results (7-10 days).

Main Results:

• The integrated SWI+CTP model achieved an AUC of 0.952, ACC of 0.909, SEN of 0.889, and SPE of 0.933.
• This performance surpassed individual SWI (AUC 0.874) and CTP (AUC 0.715) models.
• The combined model demonstrated the highest prediction accuracy.

Conclusions:

• The correlation between combined SWI and CTP features enhances prediction efficiency for infarction volume changes.
• Integrating SWI and CTP offers a superior approach for predicting cerebral infarction prognosis.
• This multimodal imaging strategy improves diagnostic accuracy in stroke management.