Advanced Brain Tumor Classification Sytem Using Deeplearning

Abstract

Diagnosing brain tumors early and accurately is a tough task in medical imaging. The challenge comes from many tumor types, their irregular shapes, and differing appearance patterns in scans. Recent advances in artificial intelligence, especially deep learning, open new doors for automatic and accurate tumor detection. This study presents a complete diagnostic system that combines convolutional neural networks (CNNs), Vision Transformers (ViTs), and ensemble techniques like Support Vector Machines (SVM) and Gradient Boosting Classifiers. To get the best input data, the system uses advanced preprocessing steps such as removing the skull, normalizing the image intensity, and correcting bias fields. Testing on common MRI datasets shows that hybrid models, especially those with transformer modules, outperform traditional models in accuracy, sensitivity, and ability to handle different tumor types. To build trust with healthcare professionals, the system includes Explainable AI features that explain how the models make decisions. These insights make it easier for doctors to understand and trust the results. Overall, the system shows strong potential to help with early diagnosis and personalized treatment of brain tumors.

Introduction

Brain tumors are life-threatening and require fast, accurate diagnosis. Traditional methods like MRI analysis can be time-consuming and error-prone. Artificial intelligence (AI)—especially deep learning (CNNs) and Vision Transformers (ViTs)—offers promising improvements in accuracy, speed, and reliability of tumor classification.

2. Objectives

The study compares three main AI approaches for brain tumor classification within a unified, interpretable framework:

• Classical machine learning (e.g., SVM, Random Forest)

• Convolutional Neural Networks (CNNs)

• Transformer-based models (e.g., ViT, Swin Transformer)

The framework also integrates data preprocessing, performance evaluation, and explainable AI (XAI) tools to ensure clinical trust and accuracy.

3. Related Work

• Traditional ML (SVM, k-NN, RF): Uses handcrafted features like texture, shape, and histogram. Effective but limited by feature quality.

• CNNs (e.g., ResNet, DenseNet): Automatically learn spatial features from MRI images. Superior in accuracy and scalability.

• Transformers (ViTs, Swin): Leverage global attention to detect irregular, fuzzy tumor boundaries. Effective in low-data settings using transfer learning.

• Explainability Tools: Techniques like Grad-CAM and SHAP build clinician trust by showing what the AI "sees" and how decisions are made.

4. Methodology

A. Data & Preprocessing

• Used BraTS dataset with multi-modal MRI scans (T1, T1c, T2, FLAIR).

• Preprocessing steps: skull stripping, bias correction, normalization, resizing, and extensive data augmentation (e.g., flips, rotations, zoom).

B. Model Types

1. Classical ML:

   • Models: SVM (RBF), Random Forest

   • Features: Texture (GLCM), shape, statistical descriptors

2. CNNs:

   • Custom architecture with convolutional, residual, and dense blocks

   • Dropout and batch norm to prevent overfitting

   • Ends with softmax classification

3. Transformers:

   • ViT and Swin Transformer, pretrained and fine-tuned on medical images

   • Adapted for grayscale inputs and small datasets

C. Training & Evaluation

• Framework used TensorFlow and PyTorch

• Dataset split: 70% training, 15% validation, 15% test

• Used Adam optimizer, categorical cross-entropy loss, and early stopping

• 5-fold cross-validation ensures robustness

D. Metrics Used

• Accuracy, Precision, Recall, F1-Score, AUC, Dice Coefficient (for segmentation)

E. Explainable AI

• Grad-CAM: Visual heatmaps for CNNs and Transformers

• SHAP: Feature-level impact for classical models

• Clinician feedback incorporated to validate interpretability

5. Results & Discussion

A. Classification Accuracy

• CNNs: ~95%+ accuracy, outperform traditional models (85–90%)

• Transformers: Match or exceed CNNs, especially with complex tumors

• Traditional ML: Lighter and easier to deploy but lower accuracy and sensitivity

B. Model Reliability

• Preprocessing and augmentation improved generalizability across MRI modalities and institutions

• Cross-validation prevented overfitting

C. Explainability

• Visual and numerical explanations made models clinically interpretable

• Alignments between AI predictions and expert radiologist feedback improved trust

D. Comparative Insights

E. Key Contributions

• A unified pipeline comparing classical ML, CNNs, and transformers

• Clinically aligned, interpretable AI models

• Benchmarked performance on open datasets (e.g., BraTS)

Conclusion

This work offers a reliable and comprehensive method for automatically identifying and classifying brain tumors. Using a hybrid approach, the suggested system combines aspects of deep convolutional networks, sophisticated transformer structures, and traditional machine learning. A thorough preprocessing pipeline enhances the quality and consistency of the MRI inputs. This pipeline includes crucial steps such as skull stripping, intensity normalization, and bias field correction. Furthermore, the study strategically employs data augmentation to improve the model\'s resilience and minimize overfitting. Through a comparative analysis of various model families, including Support Vector Machines (SVMs), Random Forests, customized Convolutional Neural Networks (CNNs), and transformer-based models like Vision Transformer (ViT) and Swin Transformer, the research demonstrates the superior performance of deep learning and transformer-based approaches. Notably, transformer models achieved classification accuracies exceeding 97%, highlighting their strong capability in capturing the spatial and contextual complexities present in brain MRI data. CNNs also produced competitive results, whereas traditional machine learning algorithms showed lower predictive power. To enhance both interpretability and clinical relevance, the study incorporates explainable AI techniques such such as SHapley and Gradient-weighted Class Activation Mapping (Grad-CAM). Additive exPlanations (SHAP). These tools provide valuable visual and quantitative insights into the model\'s predictions, thereby increasing confidence in their diagnostic validity. Future directions for this research include expanding the evaluation to encompass diverse and multi-institutional datasets, optimizing computational efficiency for real-time clinical deployment, and integrating multi-modal imaging techniques like fMRI or PET scans to further improve diagnostic accuracy and reliability.

References

[1] S. Kumar, R. Dhir, and N. Chaurasia, \"Brain Tumor Detection Analysis Using CNN: A Review,\" in Proc. 2021 Int. Conf. on Artificial Intelligence and Smart Systems (ICAIS), Coimbatore, India, 2021, pp. 1061-1067, doi: 10.1109/ICAIS50930.2021.9395920. [2] J. Islam and Y. Zhang, \"A Novel Deep Learning Based Multi-class Classification Method for Alzheimer’s Disease Detection Using Brain MRI Data,\" in Brain Informatics, BI 2017. Lecture Notes in Computer Science, vol. 10654, Springer, Cham, 2017. [3] R. Rajendran and R. Dhanasekaran, \"Fuzzy Clustering and Deformable Model for Tumor Segmentation on MRI Brain Image: A Combined Approach,\" in Proc. Int. Conf. on Communication Technology and System Design, 2011. [4] D. Aboul Dahab, S. S. A. Ghoniemy, and G. M. Selim, \"Automated Brain Tumor Detection and Identification using Image Processing and Probabilistic Neural Network Techniques,\" IJIPVC, vol. 1, no. 2, pp. 1-8, 2012. [5] T. Hossain, F. S. Shishir, M. Ashraf, M. A. Al Nasim, and F. M. Shah, \"Brain Tumor Detection Using Convolutional Neural Network,\" in Proc. 2019 1st Int. Conf. on Advances in Science, Engineering and Robotics Technology (ICASERT), Dhaka, Bangladesh, 2019, pp. 1-6, doi: 10.1109/ICASERT.2019.8934561. [6] S. S. More, M. A. Mange, M. S. Sankhe, and S. S. Sahu, \"Convolutional Neural Network Based Brain Tumor Detection,\" in Proc. 2021 5th Int. Conf. on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 2021, pp. 1532-1538, doi: 10.1109/ICICCS51141.2021.9432164. [7] K. Simonyan and A. Zisserman, \"Very Deep Convolutional Networks for Large-Scale Image Recognition,\" in Proc. of the 3rd Int. Conf. on Learning Representations (ICLR), San Diego, CA, USA, 2015, pp. 1-14. [8] X. Zhang, X. Xie, and Y. Wang, \"Brain Tumor Classification Using Deep Convolutional Neural Networks,\" in Proc. of the 2018 IEEE Int. Conf. on Bioinformatics and Biomedicine (BIBM), Madrid, Spain, 2018, pp. 905-912, doi: 10.1109/BIBM.2018.00216. [9] S. A. Khan, I. Ahmad, and H. A. Al-Rizzo, \"Brain Tumor Detection Using Convolutional Neural Networks and Transfer Learning,\" Int. J. of Imaging Systems and Technology, vol. 30, no. 5, pp. 937-944, 2020, doi: 10.1002/ima.22525.

Copyright

Copyright © 2025 Mrs. Subhashree D C, Dr. Girish Kumar D, Mr. Sai Kumar S, Mrs. Sharvani V , Ms. M M Harshitha. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.