Abstract
Objective:
This study aims to develop an AI-based radiomics model using MRI data to predict early functional outcomes in ischemic stroke patients. It evaluates the predictive accuracy of imaging-derived features compared with conventional clinical predictors. The research seeks to enhance early prognostic decision-making in stroke care.
Method:
This retrospective analytical study was conducted at a tertiary hospital in Punjab, Pakistan. The study included ischemic stroke patients who underwent MRI within 24 hours of symptom onset. Radiomic features were extracted from the MRI images, and a machine learning model was trained to predict early outcomes, specifically focusing on functional independence (as measured by the modified Rankin Scale, mRS). The model's performance was compared to that of clinical predictors, including National Institutes of Health Stroke Scale (NIHSS) scores and age.
Results:
The AI-based radiomics model demonstrated superior predictive accuracy compared to clinical predictors. The area under the receiver operating characteristic curve (AUC) for the radiomics model was 0.86, significantly higher than the AUC of 0.72 for clinical predictors alone. The radiomics features, such as lesion volume and texture heterogeneity, were identified as key indicators of functional outcomes. Additionally, integrating AI-derived MRI features with clinical predictors improved the model’s sensitivity and specificity.
Conclusion:
AI-based radiomics using MRI data offers a promising approach for the early prediction of functional outcomes in ischemic stroke patients. This model could serve as an adjunct tool for clinicians, aiding in more accurate and timely decision-making, thus improving patient management and potentially enhancing recovery prospects. The incorporation of AI into stroke prognostication could significantly influence future clinical practice and patient care pathways.