A Study on Classification and Segmentation of Brain Tumor MRI using MediAI

Abstract

Brain tumors, arising due to genetic, environmental, and immune system factors, are typically classified as primary or metastatic. Primary brain tumors originate from brain tissue and include types such as adenomas, epithelial tumors, gliomas, meningiomas, and schwannomas, while metastatic tumors spread from other body parts. The rapid and accurate diagnosis of brain tumors is crucial, and MRI is an indispensable tool in this regard due to its non-invasive nature and superior imaging capabilities. This study focuses on developing a method for the classification and segmentation of MRI images of primary brain tumors, particularly gliomas, to enhance diagnosis and treatment assessment. We propose MediAI, a Convolutional Neural Network (CNN) leveraging transfer learning with ResNet50, to classify MRI images of brain tumors. The dataset comprises 414 MRI images of primary brain tumors (86 adenomas, 84 epithelial tumors, 82 gliomas, 80 meningiomas, and 82 schwannomas) and 39 normal MR images, sourced from various public datasets. Experimental results demonstrated that MediAI achieved an impressive classification accuracy of 97.6 %. For tumor segmentation, we applied anisotropic diffusion filtering, followed by thresholding using the Otsu method, to accurately detect and delineate tumor regions, particularly in glioblastomas—highly malignant brain tumors. Tumor regions were further refined through morphological operations, and the final tumor contours were extracted and overlaid on the original images. Performance evaluation of MediAI was conducted using a confusion matrix, calculating precision, recall, and F1-score for each tumor type. Results indicated that gliomas, in particular, were c lassified with a precision of 91 %, recall of 99 %, and F1- score of 95 %. A comparison with existing studies demonstrated that MediAI outperforms previous methods, achieving the highest reported accuracy for brain tumor classification at 97.6 %. The proposed methodology not only facilitates accurate brain tumor classification but also enhances the monitoring of treatment responses by tracking changes in segmented tumor regions. Future work will focus on utilizing Radiomics to map tumor, necrotic, and edema regions for advancing diagnostic and therapeutic paradigms in glioma treatment.