Deep learning-based transfer learning approach for acute ischemic stroke classification using magnetic resonance imaging

Acute ischemic stroke (AIS) is a leading cause of mortality and morbidity worldwide, accounting for approximately 85% of all stroke cases. Early and accurate diagnosis using magnetic resonance imaging (MRI) is crucial for timely intervention and improved patient outcomes. However, manual interpretation of MRI images is time-consuming and requires specialized expertise. This study aims to develop and evaluate deep learning-based transfer learning models for automated classification of AIS using MRI images, specifically comparing the performance of VGG-16, ResNet50, InceptionV3, and VGG-19 architectures. A dataset comprising 2,400 MRI brain images (1,200 AIS cases and 1,200 normal cases) collected from multiple hospitals in Indonesia was utilized. Images were preprocessed, resized to 128×128 pixels, and augmented using rotation, zooming, shifting, and flipping techniques. The dataset was divided into 80% training, 10% validation, and 10% testing sets. Four pre-trained convolutional neural network (CNN) models were fine-tuned and evaluated using accuracy, precision, recall, and F1-score, and AUC metrics. VGG-16 achieved the highest performance with an accuracy of 96.5%, followed by InceptionV3 (94.2%), ResNet50 (93.8%), and VGG-19 (95.1%). The VGG-16 model demonstrated superior capability in feature extraction and classification, with minimal overfitting. Transfer learning with the VGG-16 architecture provides an effective and efficient approach for automated AIS classification from MRI images, offering significant potential to assist radiologists in rapid and accurate diagnosis.