Deep Learning Based Approach for Brain Tumor Detection Using MRI Images

Early diagnosis and treatment planning for patients with tumors in the brain are very essential for better management. Magnetic Resonance Imaging (MRI) is a commonly used technique for analyzing images associated with a tumor in the brain; however, manual processing for diagnosis and interpretation is a complex and highly expert-intensive procedure for radiologists, sometimes even resulting in inconsistent diagnosis. In this scenario, to overcome this demerit, this paper presents a new concept for accurate diagnosis by a deep learning-based technique for analyzing MRI images. The proposed architecture also incorporates Different types of deep learning models are used, such as Convolutional Neural Network (CNN) models for tumor classification, U-Net models, and Attention U-Net models are precise tumor segmentation. Moreover, The Super-Resolution Generative Adversarial Network (SRGAN) model is also used to increase the spatial resolution of the MRI images to capture the details within the tumor. The performance of the proposed architecture is tested using the publicly available brain MRI images. The performance measures include accuracy, Dice Coefficient, and Intersection over Union. Experimental outputs show that the developed solution is effective for tumor detection and segmentation and outperforms standard CNN and U-NET methods. In particular, the method of tumor segmentation using the attention mechanism is capable of correctly outlining boundaries of the tumor. The experiments conducted confirm the efficiency of deep-learning methods for assisting doctors and increasing the reliability of computer-assisted diagnosis systems for brain tumor analysis