Multi-Classification Brain Tumor by Mixed Transform with ResNet34

Abstract

Brain tumors pose a significant global health concern, imposing substantial social and economic burdens. Accurate classification of tumor types (gliomas, meningiomas, and pituitary tumors) from MRI data is essential for aiding radiologists and avoiding invasive biopsies. This paper presents a new method for brain tumor classification that has excellent accuracy when compared with existing methods. The proposed method provides improvements for feature extraction and classification. This new method comprises MRI image preprocessing, feature extraction, and image classification. Preprocessing includes resizing and data augmentation. Following this, features are extracted from MRI images using a mixed transform approach involving a mix of methods like DWT (Discrete Wavelet Transform), FAN (FAN Transform), and DCT (Discrete Cosine Transform). Classification employs the ResNet34 model. Where the outcomes demonstrate a training accuracy of 0.9439, a validation accuracy of 0.9195, and a best accuracy of 0.9397 when distinguishing between different brain tumor tissues in MRI images.