Outcome Assessment in Stroke Using Multiparametric MRI: Integrating Infarct Location, Radiomics, and Global Brain...

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Assessing 90-day functional outcomes in anterior circulation large vessel occlusion (LVO) stroke is challenging with conventional methods.
• Single-dimension data models have limited predictive power, necessitating multidimensional integration strategies for enhanced evaluations.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for predicting 90-day functional outcomes in LVO stroke.
• The model integrates radiomics, infarct location, brain frailty, and clinical variables for comprehensive patient assessment.

Main Methods:

• A retrospective study involving 1051 patients with anterior circulation LVO stroke across five centers (2018-2023).
• Data included T1WI, T2WI, FLAIR, and DWI MRI sequences; infarct location (Alberta Stroke Program Early CT Score); and brain frailty metrics (atrophy, WMH, old infarcts).
• Least Absolute Shrinkage and Selection Operator (LASSO) regression was employed for feature selection, with statistical analyses including AUC, decision curve analysis, and calibration curves.

Main Results:

• The integrated ML model demonstrated superior performance over single-dimension models, achieving AUCs of 0.87 (training), 0.84 (internal validation), and 0.86 (external validation).
• The fused model achieved 0.80 sensitivity and 0.80 specificity in the external validation cohort.
• Key predictive features included lentiform nucleus lesion burden, white matter hyperintensity (WMH), and M6 region lesion burden.

Conclusions:

• Integrating infarct location, brain frailty, radiomics, and clinical features significantly enhances 90-day outcome assessment in LVO stroke.
• The developed model provides an interpretable and personalized prognostic tool for clinicians managing LVO stroke patients.