Optimized transfer learning for detection susceptibility vessel sign in stroke using gorilla troops optimizer

Abstract

The blockage of blood vessels causes ischemic stroke due to clots. The susceptibility vessel sign (SVS), observed through susceptibility-weighted imaging (SWI) via magnetic resonance imaging (MRI), is a key indicator that reveals clots within brain vessels. Early detection of these clots is crucial for timely and effective treatment. Image-based detection methods, particularly non-invasive techniques like MRI, offer a superior approach compared to other modalities. This study proposes an optimized method using transfer learning to classify SVS. The deep convolutional neural network (DCNN) residual network 50 version 2 (ResNet50V2) was applied for classification, with hyperparameters fine-tuned using the gorilla troops optimizer (GTO). The optimized proposed model achieved an accuracy of 94%, sensitivity of 100%, specificity of 88%, and an F1-score of 93%. This significantly outperforms the standard ResNet50V2 model using the default parameter, which achieved an accuracy of 91%, sensitivity of 82%, specificity of 100%, and an F1-score of 77%. These results demonstrate that the proposed method significantly enhances the detection of SVS, offering a promising tool for early ischemic stroke diagnosis.