A Case for Adopting Machine Learning and Deep Learning Technologies in Kenya's Healthcare System for Brain Tumor Detection

Abstract

Brain tumors pose a dire health risk in Sub-Saharan Africa, particularly in Kenya. This research aims to develop and optimize models for detecting brain tumors in MRI images, thus minimizing diagnostic delays. The study employed a Cross Industry Standard Process for Data Mining (CRISP-DM) methodology. It utilized a dataset of 3,260 MRI images categorized into four classes: meningioma tumors, pituitary tumors, glioma tumors, and images with no tumors. Different machine learning and deep learning models, including Logistic Regression, Support Vector Machine, Decision Tree, a custom Convolutional Neural Network, and pre-trained models such as InceptionV3, VGG19, and Xception, were explored. The performance metrics revealed that deep learning models significantly outperformed machine learning models. The findings in this research underscore the need for deep learning techniques to enhance the accuracy of brain tumor classification. Hence, this elucidates their potential to reduce delays in diagnosis and improve patient outcomes. The results obtained are thus expected to contribute to the body of knowledge in early and accurate diagnosis of brain tumors in Kenya and the global medical community.