Analysis of Human Brain Tumor Lesions Based on an Enhanced ResNet-18 Model

Accurate identification of brain tumors is essential for clinical diagnosis and treatment; however, traditional manual interpretation of medical images is often inefficient and prone to inter-observer variability. To address these limitations, this paper proposes a four-class brain tumor classification method based on an improved ResNet-18 architecture, aiming to enhance the automation and accuracy of MRI image analysis. The network's ability to learn texture and structural features from MRI images is strengthened by task-specific adaptation of the input and output layers, together with transfer learning–based fine-tuning of a pre-trained model. In addition, data augmentation strategies are employed to improve the model's generalization capability. Experiments were conducted on a public brain tumor MRI dataset comprising four categories: glioma, meningioma, pituitary adenoma, and normal images. The proposed model, trained in an Apple M1 Pro environment, achieved an accuracy of 99.01% on the test set. Results from the confusion matrix demonstrate stable and consistent classification performance across all categories. Further analysis of misclassified samples reveals limitations in handling images with blurred tumor boundaries and low contrast. Overall, the proposed approach achieves a favorable balance between classification accuracy and model complexity, providing a useful technical reference for clinical decision-support systems.