Diversified AI Techniques for Augmenting Brain Tumor Diagnosis

Abstract

Brain tumors affect thousands of people worldwide each year and can be extremely fatal if not diagnosed early. They are challenging to diagnose due to their complexity and the overlapping features of different tumor types. This research explores the application of AI technology to expedite the diagnosis of brain tumors. The proposed AI-based approaches involve using deep learning algorithms to analyze medical imaging data, specifically MRI scans. The goal was to build a robust and accurate model that could overcome distribution shifts. Some of the models used include classical machine learning models and a convolutional neural network. The results demonstrate that AI-based approaches can significantly improve the accuracy and expedite the process of brain tumor diagnosis.
 The performances of the models were evaluated by using cross-validation and measuring accuracy, using a completely different dataset of MRI scans, to assess how the models performed when dealing with distribution shifts. The logistic regression model achieved a testing accuracy of 78.56%. The multi-layer perceptron (MLP) model achieved a testing accuracy of 74.89%. The multiplicative weight update method combined two models (MLP and logistic regression) with dynamically adjusted weights and achieved a testing accuracy of 83.83%. An approach where multiple aggregating models were used collaboratively achieved a testing accuracy of 86.46%. The best performing model, a convolutional neural network (CNN), yielded a testing accuracy of 98.20%.