Abstract Background: Acute ischemic stroke is a leading cause of death and long-term disability worldwide, with a disproportionately increasing burden in low- and middle-income countries. Diffusion-weighted imaging (DWI) is the most sensitive magnetic resonance imaging (MRI) sequence for the early detection of acute ischemia, as treatment decisions, including intravenous thrombolysis and mechanical thrombectomy, are highly time-dependent. However, interpretation of subtle lesions may be challenging, particularly in high-volume emergency settings, potentially leading to diagnostic delays. Objective: To train and test a DWI-based deep learning model for the automated detection of supratentorial acute ischemic infarcts and assess its diagnostic performance relative to expert radiologist interpretation. Materials and Methods: This retrospective diagnostic accuracy study included adult patients who were suspected of having acute ischemic stroke and underwent an MRI the brain in a tertiary care teaching hospital. Only the supratentorial DWI images were analyzed. The studies were classified into normal or infarct-positive according to consensus interpretation of the experienced radiologists with correlation to apparent diffusion coefficient (ADC) maps. The deep learning-based classification model with 70:15:15 train–validation–test split was trained and tested using images that had undergone preprocessing. Model performance was assessed using accuracy, sensitivity, specificity, F1-score, and area under the curve (AUC). Results: A total of 1024 MRI studies were included. The proposed model showed high diagnostic accuracy on the independent test data, with an overall accuracy of 90.8%, a sensitivity of 92.1%, a specificity of 89.6%, and an AUC of 0.93. The average processing time per study was <1 s. Conclusion: The DWI-based deep learning model demonstrated high diagnostic accuracy in the identification of supratentorial acute ischemic infarcts. This system may serve as a decision-support tool to assist radiologists and enhance workflow efficiency, particularly in time-sensitive and resource-constrained settings.
Katam et al. (Thu,) studied this question.