AI-Enabled Stroke Detector: A Hybrid Deep Learning and Machine Learning Approach

Stroke is a significant cause of fatalities and permanent disability globally, and its diagnosis in resource-limited settings is typically affected by the necessity to have a specialist to interpret CT scans. To address this, a hybrid diagnostic solution combining machine learning and deep learning is proposed. Preprocessing of CT brain scans of healthy individuals and stroke patients was done using Albumentations (resizing, normalization, augmentation). Various models were evaluated with 5-fold crossvalidation including Logistic Regression, Random Forest, Decision Tree, CNN, MobileNet, ResNet50, InceptionNet, and a hybrid CNN+DNN model. Random Forest and InceptionNet achieved the highest performance, with both having a test accuracy of 97.92 %, with InceptionNet demonstrating excellent feature extraction in its inception modules. A GUI was used to deploy the trained models and support rapid stroke recognition. This approach provides practical and clinical utility to improve stroke diagnosis and accelerate interventions in low-resource healthcare settings.