Fine-Tuned Xception Model for Accurate Brain Tumor Classification: A Deep Learning Approach

The conditions are fatal and therefore the need for early diagnosis and accurate results to enhance on proper treatment. The modern advances in the field of Deep learning show great promises in automating the analysis of medical images and bringing accuracy to diagnostic levels. In this work, the performance effectiveness of the fine-tuned Xception model for classifying brain tumors based on MRI images is investigated. Using glioma, meningioma, pituitary tumor, the Xception model was fine-tuned as a transfer learning strategy to identify appropriate higher-level features that increases classification quality. To make the model less sensitive to both the image size and its orientation additional preprocessing steps were used: resizing, normalization and data augmentation. The proposed model provided approximately 96% classification accuracy, and it was faster than the standard CNN architectures. While precision, recall, and accuracy, as well as the F1-score, provided better evidence across all classes of the model's accuracy. The approach used in the current work demonstrates the effectiveness of using the Xception model in the diagnosis of brain tumors and the need to reduce the manual interpretation process in clinics. Future work also plans to use the framework in real-time diagnostic systems where a broader application of better medical imaging will be more available to the public.