Related Experiment Video
Updated: Jul 20, 2026
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
43.8K
Computerized diagnosis of brain tumor using graph based CNN classification
C Agees Kumar1, P V Deepa2, T S Sivarani1
1Dept. of EEE, Arunachala College of Engineering for Women, Vellichanthai, Nagercoil, Kanyakumari District, Tamil Nadu, 629203, India.
Computers in Biology and Medicine
|April 9, 2026
Summary
This study introduces an automated brain tumor classification framework using MRI images. The novel approach combines Graph Neural Networks and Convolutional Neural Networks for high accuracy in distinguishing tumor types.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Oncology
- Computational Pathology
Background:
- Accurate brain tumor diagnosis via MRI is crucial for treatment planning but relies on time-consuming subjective expert interpretation.
- Existing computer-aided diagnosis methods using Convolutional Neural Networks (CNNs) often fail to capture complex tumor structures and heterogeneity.
- Graph-based methods struggle to integrate segmentation-based tumor structures within a unified learning framework.
Purpose of the Study:
- To develop a robust, automated four-stage classification framework for brain tumors using MRI images.
- To overcome the limitations of traditional CNNs and graph-based methods in representing tumor structural associations and heterogeneity.
- To enhance the discrimination between benign and malignant tumors by incorporating inter-region dependencies and local texture cues.
Main Methods:
- The proposed framework utilizes adaptive bilateral filtering for noise reduction and edge enhancement, followed by semantic segmentation to isolate tumor regions.
- Structural boundary information is encoded using Histogram of Oriented Gradients (HOG) descriptors.
- A hybrid Graph Neural Network-Convolutional Neural Network (GNN-CNN) model learns region-level representations as graph nodes, capturing spatial and feature similarities.
Main Results:
- The framework achieved high performance on the Kaggle brain tumor dataset.
- Experimental results demonstrated 98.5% accuracy, 98.5% precision, 97.8% recall, and 97.8% F1-score.
- The method effectively integrates inter-region dependencies and local texture cues for improved tumor classification.
Conclusions:
- The proposed GNN-CNN framework offers an efficient and high-performing automated solution for brain tumor classification from MRI data.
- The integration of graph-based and convolutional approaches enhances the model's ability to capture complex tumor characteristics.
- This automated approach has the potential to significantly aid clinical decision-making in neuro-oncology.
