Brain Tumor Classification With Selective Fine Tuning Using Transfer Learning

摘要

The accelerated pace of contemporary life has led to a notable increase in cancer incidence, which poses a significant challenge in the field of oncology. This study introduces an innovative approach to brain tumor detection by employing fine-tuned pre-trained models with sparse data and comparing their performance to that of traditional convolutional neural networks (CNNs). The study addresses the challenge of limited medical imaging datasets in oncology, a discipline experiencing heightened demand due to rising cancer rates. By utilizing transfer learning techniques, the proposed method seeks to alleviate the overfitting issues that are commonly encountered. Fine-tuned models developed from pre-trained networks exposed to millions of diverse images have been adapted for tumor classification tasks by incorporating max-pooling and dense layers. A comparative analysis revealed that these refined models achieved superior accuracy, exceeding 90 percent even with limited data, thereby outperforming the conventional CNNs. This study evaluated the model performance using various metrics, including accuracy and precision, and demonstrated the efficacy of transfer learning in enhancing brain tumor detection capabilities. This approach holds promise for improving diagnostic tools in oncology, particularly in scenarios in which large-scale medical imaging datasets are unavailable.