Transfer Learning in Convolution Neural Network for Brain Tumor Detection Using a Small Training Dataset

Abstract

Tumors are a significant risk in today's medical field, requiring fast and reliable automated techniques for detection, particularly for brain tumors. Accurate detection is crucial for effective treatment and saving lives. Various image processing techniques aid doctors in providing appropriate treatment. Manual or human-based identification of brain tumors using MRI images is time-consuming and prone to inaccuracies, especially for an experienced person. Deep learning algorithms have introduced effective solutions for brain tumor detection. The one constraint is that the algorithms need to train on a huge amount of data for reliable performance. This research aims to investigate the effectiveness of transfer learning for brain tumor detection using a small training dataset regardless of the tumor type. Therefore, we have used three models of Convolutional Neural Networks (CNN): traditional model, enhanced model, and transfer learning-based model. The average results have shown that the transfer learning-based model has better performance on the small training dataset than traditional and enhanced models, with a classification accuracy reached 92%.