Multi-scale DWI-based Radiomics for predicting infarct volume evolution in acute ischemic stroke beyond the...

Area of Science:

• Medical Imaging
• Radiology
• Neurology

Background:

• Acute ischemic stroke (AIS) management often extends beyond the traditional therapeutic window.
• Predicting infarct evolution and neurological outcomes in these late-window AIS patients is crucial for guiding treatment decisions.

Purpose of the Study:

• To investigate the predictive capability of radiomics features derived from multi-scale diffusion-weighted imaging (DWI) for infarct volume reduction in subacute AIS patients.
• To assess the correlation between predicted infarct volume changes and clinical neurological function.

Main Methods:

• Retrospective analysis of 72 AIS patients (Center 1) and 29 (Center 2).
• Delineation of core infarct volumes and peri-core regions (2-mm, 5-mm) on baseline DWI and follow-up FLAIR imaging.
• Extraction and selection of radiomics features using LASSO regression, followed by model building (logistic regression, XGBoost, LightGBM, random forest) with cross-validation and external validation.

Main Results:

• The radiomics model using the 2-mm peri-core region achieved the best performance (cross-validated AUC = 0.827 ± 0.064; external AUC = 0.804).
• This radiomics model significantly outperformed a clinical model (AUC = 0.726 ± 0.129).
• NIHSS score change significantly correlated with infarct volume change (ρ = 0.291, p = 0.003), with improved neurological function observed in 63.4% of the volume-improvement group.

Conclusions:

• Radiomics features from the 2-mm peri-core region of DWI provide robust prognostic information for infarct evolution in late-window AIS.
• These imaging-derived features demonstrate a significant correlation with clinical neurological function, aiding in outcome prediction.