<![CDATA[The prevalence of ischemic stroke in Indonesia continues to increase, highlighting the need for rapid and accurate diagnostic approaches. Computed Tomography (CT) imaging plays a crucial role in the early detection of ischemic stroke, particularly within the therapeutic window of 3–4.5 hours after symptom onset. This study aimed to develop and evaluate a website-based ischemic stroke detection system using a MobileNet deep learning model integrated with CT scan image analysis. A Research and Development (R&D) approach employing the Rapid Application Development (RAD) framework was applied. The model was developed using 2,000 CT scan images obtained from a Kaggle dataset, comprising 1,000 ischemic stroke images and 1,000 normal control images. The dataset was divided into training and validation sets, with 800 images per class used for training and 200 images per class used for validation. External clinical validation was conducted using 600 CT scan images collected from hospitals, consisting of 300 ischemic stroke images and 300 normal control images. Model performance was evaluated using accuracy, precision, sensitivity, and specificity metrics, while system usability was assessed using the System Usability Scale (SUS). The developed model achieved an accuracy of 92.25%, precision of 92.00%, sensitivity of 92.50%, and specificity of 92.00% on the validation dataset. However, performance decreased during external clinical testing, yielding an accuracy of 71.50%, precision of 73.12%, sensitivity of 68.00%, and specificity of 75.00%. These findings are consistent with previous studies indicating that models trained solely on public datasets often experience performance degradation when applied to real-world clinical data, emphasizing the importance of multi-institutional training datasets and extensive external validation. The usability evaluation produced an average SUS score of 83, indicating excellent user acceptance and system usability. The proposed website-based MobileNet model demonstrates strong potential as an accessible tool for supporting early ischemic stroke detection from CT scan images and may be further enhanced through the incorporation of multi-slice or volumetric imaging data to improve clinical performance.]]>
