📄 IEEE Published Research | IC-CGU 2025
A machine learning based system for brain tumor classification using MRI images by combining Local Binary Pattern (LBP), Image Subtraction Analysis, and k-Nearest Neighbors (k-NN).
Paper Title
Enhancing Brain Tumor Classification Through Image Subtraction Analysis
Conference
IEEE International Conference on Cognitive, Green and Ubiquitous Computing (IC-CGU 2025)
📑 IEEE Paper Link
https://ieeexplore.ieee.org/abstract/document/11338093
Brain tumor diagnosis using MRI images is a critical task in medical analysis. Traditional texture extraction methods such as Local Binary Pattern (LBP) are sensitive to noise and may fail to capture complex tumor textures.
This research proposes an improved feature extraction method using image subtraction to enhance tumor textures and improve classification accuracy.
The proposed system works by:
- Extracting texture features using LBP
- Subtracting the original MRI image from the LBP image
- Enhancing tumor edges and texture patterns
- Feeding the processed image features into a k-NN classifier
This approach reduces noise and improves tumor feature representation.
The model classifies MRI images into three tumor types:
- Glioma
- Meningioma
- Pituitary
MRI Image
↓
Grayscale Conversion
↓
Local Binary Pattern (LBP)
↓
Image Subtraction (LBP − Original Image)
↓
Image Resizing (64×64)
↓
Normalization
↓
Feature Extraction
↓
K-Nearest Neighbors (KNN)
↓
Brain Tumor Classification
Two publicly available datasets were used:
- Figshare Brain MRI Dataset
- Kaggle Brain Tumor MRI Dataset
Dataset Characteristics:
- Image size: 512 × 512
- Classes: Glioma, Meningioma, Pituitary
- Training split: 80%
- Testing split: 20%
The proposed model achieved strong classification performance.
| Metric | Value |
|---|---|
| Accuracy | 95.44% |
| Precision | High |
| Recall | High |
| F1 Score | Balanced |
The following figures illustrate the visual outcomes of the proposed brain tumor classification approach.
These include the original MRI images, LBP-transformed images, texture-differentiated images, and their corresponding histograms.
This visualization demonstrates the transformation pipeline:
- Original MRI Image
- LBP Feature Extraction
- Texture Differentiation using Image Subtraction
- Histogram Analysis of Pixel Distribution
)
These histograms represent the pixel intensity distribution after texture differentiation, which helps in distinguishing between different tumor types.
This project introduces a novel feature enhancement technique for brain tumor classification by combining Local Binary Pattern (LBP) with image subtraction analysis.
The proposed method improves traditional LBP-based approaches by:
- Reducing noise in MRI images
- Enhancing tumor edges and textures
- Improving classification accuracy
- Achieving 95.44% accuracy in tumor classification
This research demonstrates the potential of combining image processing and machine learning techniques for improved medical image analysis.
- Python
- OpenCV
- NumPy
- Scikit-learn
- Image Processing
- Machine Learning
Future work may include:
- Deep Learning models (CNN)
- Transfer Learning
- Data Augmentation
- Hybrid ML + DL architectures
- Real-time medical diagnostic systems
Manas Ranjan Bhui
B.Tech – Computer Science and Engineering
CV Raman Global University
Skills
- Machine Learning
- Data Analytics
- Python
- SQL
- Power BI
- Computer Vision
If you use this work, please cite:
M. R. Mishra, J. Nayak, A. A. Tripathy, S. Jena, M. R. Bhui, and P. K. Meher.
Enhancing Brain Tumor Classification Through Image Subtraction Analysis.
IEEE International Conference on Cognitive, Green and Ubiquitous Computing (IC-CGU), 2025.
DOI: https://ieeexplore.ieee.org/abstract/document/11338093
⭐ If you found this project useful, please consider giving it a star!