Machine learning and transfer learning techniques for accurate brain tumor classification

Abstract

Brain tumors, resulting from uncontrolled and rapid cell growth, pose significant health risks if not treated early. Despite numerous advancements, accurate segmentation and classification remain challenging. This study leverages machine learning (ML) and transfer learning techniques to classify healthy and sick individuals using numerical data and MRI images. We utilized 3762 MRI images alongside Light Gradient Boosting Machine (LightGBM), AdaBoost, gradient boosting, Random Forest, Quadratic Discriminant Analysis, Linear Discriminant Analysis, logistic regression, and transfer learning algorithms. Numerical data was processed with LightGBM, achieving an accuracy of 95.7 %. Transfer learning applied to image data using a modified GoogLeNet model further enhanced classification accuracy to 99.3 %. These results demonstrate the effectiveness of combining ML and transfer learning techniques for accurate brain tumor classification, addressing limitations of prior approaches and offering improved diagnostic reliability. All coding and model implementations were conducted on the Python platform.