An Automated Brain Tumor Detection and Prediction in MRI images using Deep Learning Algorithms

Read full text Summarise with VIRAT
Authors: Tatapudi Ravi Raju, V. Manikanta, P.K. Steve Waugh, S.V.N. Sai Charan, V. Hari Krishna Prasad
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Brain tumor detection plays a critical role in medical diagnosis, as early and accurate identification greatly improves treatment planning and patient survival. Magnetic Resonance Imaging (MRI) is widely used for tumor analysis because it is non-invasive and provides high-resolution images, yet manual interpretation by radiologists can be time- consuming and subject to variability. Traditional detection systems relied on handcrafted feature extraction combined with machine learning classifiers such as SVM, KNN, and Random Forest. More recent approaches employ deep learning and transfer learning models like VGG16, ResNet50, and InceptionV3, along with attention mechanisms and ensemble strategies to enhance performance. However, these systems still face challenges including overfitting, dataset dependency, limited segmentation accuracy, and computational complexity, resulting in classification accuracies typically ranging from 60% to 75%, which is insufficient for reliable clinical deployment. To address these limitations, the proposed system introduces an automated deep learning framework for brain tumor detection and classification using MRI images. Advanced preprocessing techniques—including filtering, normalization, and frequency-domain transformations—enhance image quality, while segmentation isolates tumor regions for focused analysis. A Convolutional Recurrent Neural Network (CRNN) combines CNN-based spatial feature extraction with RNN-based contextual learning, further strengthened by an attention mechanism to highlight tumor- relevant features. Ensemble classification improves robustness and prediction reliability. Evaluated using standard metrics such as accuracy, precision, recall, and F1-score, the proposed framework achieves approximately 95% classification accuracy across tumor categories including glioma, meningioma, pituitary, and no tumor. This integrated system provides a fast, objective, and consistent diagnostic tool that reduces radiologist workload while improving detection accuracy, with future enhancements aimed at multimodal MRI integration and explainable AI to increase clinical trust. Furthermore, a Flask-based web application is implemented to deploy the proposed model in a practical and accessible environment. The application provides an interactive interface where users can upload MRI images for analysis. The backend integrates the trained CRNN model using a Keras (.h5) file, while image preprocessing is handled using OpenCV and NumPy. The system performs real-time prediction and displays the classification result (Glioma, Meningioma, Pituitary, or No Tumor) along with confidence scores and processing time. Designed with HTML, CSS, and JavaScript for the frontend, the application ensures responsive performance across devices and supports multiple concurrent requests. This deployment enhances usability by enabling fast, reliable, and user-friendly brain tumor detection outside traditional clinical settings.

Fields of Study & Subjects

Brain Tumor Detection and Classification AI in cancer detection Medical Image Segmentation Techniques

Publication details

Journal / source
International Journal for Modern Trends in Science and Technology
Publication date
21 March 2026
Language
EN