Development of an Efficient Stacking-Based Machine Learning Framework for Brain Stroke Diagnosis Using CT Images

Abstract

Brain stroke is a severe pathological condition caused by interrupted blood flow to the brain, resulting in possible neurological damage or death. Accurate and timely diagnosis through Computed Tomography (CT) scan images is a prerequisite for immediate medical intervention. The current research is proposing an effective stacking-based machine learning framework for brain stroke diagnosis automatically using feature extraction, feature selection, dimensionality reduction, and ensemble learning methods. The method starts with CT image preprocessing to convert images into grayscale, downsize them into a uniform 224×224 resolution, and normalize pixel intensity. Feature extraction is done for texture-based statistical features using Gray Level Co-occurrence Matrix (GLCM) and Local Binary Patterns (LBP) for texture pattern description. For improved model performance, SelectKBest (ANOVA F-test) is utilized for the most discriminative feature selection, and Principal Component Analysis (PCA) for dimensionality reduction. Synthetic Minority Over-sampling Technique (SMOTE) is utilized to address class imbalance and enhance generalization.A stacking ensemble learning model is built for classification purposes, with Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) as base learners and XGBoost as the meta-learner. Hyperparameter tuning is carried out using GridSearchCV for model performance optimization. The results of the experiment, carried out on a labeled CT image dataset, show high accuracy at 96.21%, with precision at 0.95 (Normal) and 0.98 (Stroke), recall at 0.99 (Normal) and 0.92 (Stroke), and F1-score at 0.97 (Normal) and 0.95 (Stroke). These findings indicate that the proposed framework significantly improves stroke classification accuracy compared to conventional machine learning models.This research illustrates the potential of stacking-based ensemble learning in medical imaging and provides an efficient method for automatic, non-invasive, and accurate diagnosis of stroke. The proposed system has the ability to assist radiologists and healthcare professionals in achieving quick and accurate decisions, subsequently leading to improved patient outcomes.