Optimized brain tumor analysis in FLAIR-MRI LGG images: leveraging transfer learning and optimization for enhanced diagnosis and localization

This research endeavour conducts a comprehensive exploration of an efficient approach for categorizing and delineating brain tumors in fluid-attenuated inversion recovery magnetic resonance imaging (FLAIR-MRI) low grade glioma (LGG) images. The precise diagnosis and localization of brain tumors are pivotal tasks in the domain of medical imaging, and the proposed approach demonstrates notable advancements in both accuracy and computational efficiency. In the realm of classification, the investigation harnesses the formidable residual network-50 (ResNet-50) architecture by applying transfer learning techniques. Transfer learning facilitates the utilization of pre-trained neural network weights, significantly augmenting the model's ability to generalize from a relatively limited medical image dataset. To further refine the classification model, the grey wolf optimizer (GWO), a biologically inspired optimization algorithm, was employed. This strategic choice enabled meticulous fine-tuning of vital parameters, including learning rate, dense layer configurations, batch size, and the number of training epochs. Subsequently, for the segmentation task, the previously developed ResUNet model was leveraged, specifically tailored for brain tumor segmentation. By seamlessly integrating this model with the classification framework, a comprehensive solution for brain tumor analysis is presented, simultaneously delivering an accurate diagnosis and precise localization of tumor regions within the images. The resulting classification model exhibits remarkable performance metrics, achieving a testing accuracy of 90.7%, a sensitivity of 88.8%, and a specificity of 97.01%. These metrics indicate the model's proficiency in discerning tumor and non-tumor regions within FLAIR-MRI images. This research highlights the potential of advanced deep learning (DL) techniques and optimization strategies to enhance the efficacy and reliability of brain tumor analysis in medical imaging, ultimately aiming to improve patient care and treatment planning.