Neuro-VGNB: Transfer Learning-Based Approach for Detecting Brain Stroke

Abstract

A brain stroke occurs when blood flow to the brain is interrupted leading to potential brain damage and loss of functions controlled by the affected area. Timely diagnosis and intervention are critical for minimizing long-term disabilities and improving recovery outcomes.This study addresses the critical challenge of brain stroke detection by utilizing a combination of deep learning and machine learning techniques. We first extracted features from the VGG16 model, a well-established convolutional neural network known for its efficacy in image classification tasks. These extracted features were then enhanced and transferred using the Gaussian Naive Bayes (GNB) model, integrated with non-negative matrix factorization to optimize feature representation. Our innovative approach, termed Neuro-VGNB, aims to leverage these advanced methodologies to improve classification accuracy significantly.To evaluate the effectiveness of our Neuro-VGNB model, we conducted comprehensive comparisons between traditional spatial features and the newly proposed transfer features. Remarkably, the Logistic Regression (LR) model yielded a high accuracy score of 99.96% indicating the robustness of our method. Furthermore, we employed k-fold cross-validation to ensure reliable performance assessment and to facilitate state-of-the-art comparisons with existing methods in the literature. The findings of this study not only highlight the potential of our Neuro-VGNB approach in enhancing the detection of brain strokes but also demonstrate its applicability in clinical settings. Our results suggest that integrating advanced deep learning techniques with machine learning can significantly improve diagnostic accuracy paving the way for more effective stroke detection systems.

Keywords

• Brain stroke
• machine learning
• transfer learning
• Neuro-VGNB and VGG-16