Brain Tumor Classification using Semi Supervised Self-Training Algorithm

No Thumbnail Available

Date

2022-12

Journal Title

Journal ISSN

Volume Title

Publisher

International Islamic University Chittagong

Abstract

Brain tumors have been considered the world’s most dangerous medical condition. Worldwide, approximately 0.25 million people die every year due to CNS tumors and primary cancerous brains. The existing method is obtrusive, tedious, and sensitive to errors by individuals. To overcome the pitfalls mentioned above for brain tumor multi class classification, an approach named confidence regularized self training for brain tumor classification (CRST-BTC) has been proposed for the early detection of brain tumors in an efficient way. For the purpose of early diagnosis, this research introduces a novel approach to enhance the accuracy of brain tumor classification using confidence regularization self-training. The study systematically investigates the impact of varying percentages of unlabeled data and explores different backbone networks, with VGG16 emerging as the standout performer. Additional improvements, including integrating a dense layer and optimization algorithms such as Adam, contribute to superior clas sification accuracy. Surprisingly, the model performs better with 20% labeled data, challenging conventional expectations. The research leverages the SHAP framework for visualization, providing insights into feature importance. Benchmarking against state of-the-art models showcases the competitive edge of the proposed methodology. The VGG16 backbone with the Adam optimizer stands out, achieving 97% accuracy, 98.67% precision, 97.45% recall, and 98.01% F1-score. This study advances the field and pro vides valuable insights for practitioners in medical image analysis

Description

This Dissertation is Submitted in Fulfillment of the Requirements for the Degree of Bachelor of Science (B.Sc.) in Computer Science and Engineering (CSE) Autumn-2022

Keywords

Brain Tumor, Self-Training, Confidence Regularization, SHAP

Citation