Brain Tumor State Prediction Using Image Segmentation Techniques

Due to the incidence of brain tumors-both slow-growing and fast-growing types-occurring at approximately 18.71 cases per 100,000 people worldwide, their significance has become increasingly apparent. Fast-growing tumors, in particular, complicate treatment, increase risks, and adversely affect both physical and mental health while raising mortality rates. Therefore, this thesis focuses on developing a program for brain tumor staging classification using deep learning technology. The Convolutional Neural Network (CNN) architecture was chosen as the primary model due to its high efficiency in processing medical images and relatively short training time. The developed system was trained and tested using an MRI brain tumor image dataset, divided into training, validation, and testing sets. Experimental results showed that the model achieved a mean IoU (Intersection over Union) score of 0.885, indicating the overlap between the model's output and the ground truth, and a classification accuracy of 98.71 % for brain tumor staging (averaged across all classes).