Development of a web application for stroke diagnosis assistance using deep learning artificial intelligence on computed tomography image

Abstract

Stroke presents a significant health risk for the elderly, demanding precise diagnosis through computed tomography (CT) imaging. Expert interpretation is crucial, and present artificial intelligence (AI) proves invaluable for radiologists, ensuring accuracy. Various accessible programs, but requiring payment, can be installed on devices. This study aims to develop a web application for stroke detection in brain CT images. The web application, compatible with computers, phones, and tablets, employs deep learning AI for stroke classification. Using a dataset of 1,636 images (1,111 normal, 525 stroke), about 70% (1,175 images) were used to train, 20% (329 images) for validation, and 10% (132 images) for test the AI model (Deep learning: VGG-16). Evaluation metrics, including accuracy, sensitivity, specificity, F1-score, and are under curve (AUC), gauged the web app's performance. Design and functionality were assessed through a 5-point Likert scale by three radiologists. Results show impressive accuracy, sensitivity, specificity, F1-score, and AUC (0.969, 0.952, 0.978, 0.952, 0.965). Design and performance scores were 4.13 ± 0.38 and 4.37 ± 0.33. In conclusion, the web application effectively diagnoses strokes in brain CT images, offering a user-friendly experience on the internet.