Brain Stroke Detection in CT Scan Images Using an Enhanced Reduce Dimensionality Pattern-based CNN (ERDP-CNN) Model

Stroke is a disorder resulting from insufficient blood flow to the brain, and needs to be diagnosed as soon as possible to be treated effectively and to improve patient outcomes. Examining CT images for stroke detection is challenging, given the tight time constraints and the heightened risk of diagnostic errors under pressure. The proposed work resolves these challenges and introduces a new model named an Enhanced Reduce Dimensionality Pattern Convolutional Neural Networks (ERDP-CNN) to improve stroke detection accuracy and efficiency in brain CT images. Using a dataset from Kaggle with labelled CT scans for 2,500 stroke cases and 2,500 non-stroke cases (each image sized 650 × 650 pixels in grayscale). The research has implemented K-means clustering to identify Regions of Interest (ROIs), followed by the Local Binary Pattern (LBP) method to extract reliable features. After the feature extraction, a feature selection method was developed using the Whale Optimization Algorithm (WOA)—the CNN classifier's outcomes were enhanced by these techniques. The ERDP-CNN model was developed to reduce data dimensionality while optimising feature analysis and classification. The proposed model is evaluated using key metrics—Accuracy, F1-score, precision, specificity, and sensitivity, demonstrated that ERDP-CNN offers a superior stroke detection accuracy rate achieved by 98.97 % and compared to traditional methods like SVM, OzNet, highlighting its potential to improve diagnostic speed and reliability in clinical practice.