abink / Tumor_Classification

Brain MRI Tumor Classification Using Deep Learning

1. Introduction

Cancer remains one of the major global health challenges, and the increasing availability of medical imaging has created opportunities to support clinicians through computer-aided image analysis. In oncology, medical imaging plays an important role in the detection, characterization, treatment planning, and follow-up of tumors.

Brain MRI is particularly important in the assessment of brain tumors because MRI provides detailed soft-tissue contrast without exposing the patient to ionizing radiation. However, interpretation of large numbers of medical images can be time-consuming and requires considerable clinical expertise.

This project explores the use of deep learning for automated brain MRI tumor classification , combining concepts from medical imaging, artificial intelligence, machine learning, and deep learning .

The objective was not simply to train a CNN and report its accuracy, but to investigate how different pretrained deep-learning architectures perform on brain MRI data and how transfer learning, layer selection, fine-tuning, optimization, and regularization affect classification performance.

---

2. Medical Imaging Dataset

The project uses brain MRI images representing four classification categories:

1. Glioma
2. Meningioma
3. Pituitary tumor
4. No Tumor

Gliomas, meningiomas, and pituitary tumors represent three major tumor categories considered in this classification task, while the No Tumor class provides a non-tumor reference category.

The dataset was organized as a four-class image-classification problem with a balanced test set of 1,600 images (400 per class) .

A consistent preprocessing and augmentation pipeline was maintained across the different architectures to make the model comparison more meaningful.

Input image size: 128 × 128 × 3

---

3. Project Objective

The main objective was to investigate whether transfer-learning-based CNN architectures could effectively learn discriminative features from brain MRI images and determine which architecture provided the most reliable classification performance.

Rather than relying on a single neural network, multiple established architectures were evaluated:

• VGG16
• ResNet50
• DenseNet121
• EfficientNetB0

The models were first evaluated using their pretrained ImageNet features and subsequently investigated through domain-specific fine-tuning .

The project also examined how changing the classification head, trainable layers, learning rate, optimizer, regularization, and fine-tuning strategy affected performance.

---

4. Transfer Learning and Model Development

All models used ImageNet-pretrained convolutional backbones . The original classification layers were removed and replaced with task-specific classification heads.

The general workflow was:

Brain MRI Dataset
        ↓
Preprocessing & Augmentation
        ↓
Train / Validation / Test Split
        ↓
ImageNet-Pretrained CNN
        ↓
Custom Classification Head
        ↓
Initial Transfer Learning
        ↓
Layer Unfreezing & Fine-Tuning
        ↓
Performance Evaluation
        ↓
Model Comparison

For the classification heads, architectures using:

GlobalAveragePooling2D
        ↓
Dense Layers
        ↓
Batch Normalization
        ↓
Dropout
        ↓
4-Class Softmax

were investigated.

Fine-tuning was performed by selectively unfreezing deeper convolutional layers while keeping earlier feature-extraction layers frozen. Lower learning rates were used during fine-tuning to avoid excessively changing useful pretrained features.

---

5. Model Comparison

The four architectures showed substantially different behavior on the brain MRI classification task.

Best-performing model

Fine-tuned VGG16

• Test Accuracy: 91.50%
• Macro F1: 0.91
• Weighted F1: 0.91

Fine-tuning improved VGG16 from 83.44% to 91.50% , an improvement of 8.06 percentage points .

---

6. VGG16

VGG16 was the strongest architecture evaluated in this project.

Classification Architecture

VGG16 (ImageNet pretrained)
        ↓
GlobalAveragePooling2D
        ↓
Dense(256) + BatchNorm + Dropout(0.4)
        ↓
Dense(128) + BatchNorm + Dropout(0.3)
        ↓
Dense(4, Softmax)

The initial VGG16 experiment used the pretrained backbone with a custom classification head.

The classification head was subsequently improved by replacing Flatten() with GlobalAveragePooling2D , reducing the number of parameters and helping control overfitting.

Fine-Tuning

The final convolutional block was unfrozen while earlier layers remained frozen.

Training used:

• Adam optimizer
• Initial learning rate: 1 × 10⁻⁴
• Fine-tuning learning rate: 1 × 10⁻⁵
• Early stopping
• ReduceLROnPlateau
• Model checkpointing

Performance

Final accuracy: 91.50%

The strong performance across the tumor classes made VGG16 the selected architecture for subsequent stages of the project.

---

7. ResNet50

ResNet50 was evaluated as an alternative deep residual architecture.

Architecture

ResNet50 (ImageNet pretrained)
        ↓
GlobalAveragePooling2D
        ↓
Dense(256) + BatchNorm + Dropout(0.4)
        ↓
Dense(128) + BatchNorm + Dropout(0.3)
        ↓
Dense(4, Softmax)

The backbone was initially frozen and the classification head was trained. The final convolutional section was subsequently unfrozen for domain-specific fine-tuning.

Results

Initial accuracy: 68.00%

Fine-tuned accuracy: 81.00%

Fine-tuning produced a substantial improvement of approximately 13 percentage points , demonstrating that adaptation of pretrained features to the MRI domain was beneficial.

However, ResNet50 remained below the performance of fine-tuned VGG16.

---

8. DenseNet121

DenseNet121 was included because dense feature reuse and hierarchical feature propagation make it an important architecture for medical-image classification.

Architecture

DenseNet121 (ImageNet pretrained)
        ↓
GlobalAveragePooling2D
        ↓
Dense(256) + BatchNorm + Dropout(0.4)
        ↓
Dense(128) + BatchNorm + Dropout(0.3)
        ↓
Dense(4, Softmax)

Fine-Tuning Investigation

The initial fine-tuning strategy reduced performance from:

83% → 79%

A more conservative DenseNet-specific strategy was then investigated:

• Later convolutional layers selectively unfrozen
• Batch Normalization layers kept frozen
• AdamW optimizer
• Learning rate: 3 × 10⁻⁶
• Weight decay: 1 × 10⁻⁴
• Early stopping
• ReduceLROnPlateau

The optimized configuration recovered the performance to 83% , but did not provide a meaningful overall improvement over the original frozen-backbone configuration.

Best Results

Accuracy: 83.00%

Macro F1: 0.82

DenseNet121 therefore provided a strong baseline but did not outperform the fine-tuned VGG16.

---

9. EfficientNetB0

EfficientNetB0 was evaluated using the same dataset and general transfer-learning framework.

Architecture

EfficientNetB0 (ImageNet pretrained)
        ↓
GlobalAveragePooling2D
        ↓
Dense(256) + BatchNorm + Dropout(0.4)
        ↓
Dense(128) + BatchNorm + Dropout(0.3)
        ↓
Dense(4, Softmax)

The backbone was initially frozen and subsequently partially fine-tuned.

Results

Initial accuracy: 33%

Fine-tuned accuracy: 35%

The model exhibited severe class-collapse behavior and failed to reliably identify some of the tumor categories.

Therefore, EfficientNetB0 was not selected for further development within this project.

---

10. Overall Findings

The experiments demonstrated that architecture selection and fine-tuning strategy have a major effect on medical-image classification performance .

The results were:

An important observation was that fine-tuning did not improve every architecture equally .

• VGG16 responded strongly to domain-specific fine-tuning.
• ResNet50 showed substantial improvement.
• DenseNet121 required a more conservative fine-tuning strategy and ultimately remained around its baseline performance.
• EfficientNetB0 showed severe classification difficulties under the tested configuration.

This demonstrates that pretrained architectures cannot simply be treated as interchangeable models; their layer structure, feature representations, and fine-tuning behavior need to be considered when applying deep learning to medical images.

---

11. Final Model

Based on the comparative evaluation, fine-tuned VGG16 was selected as the best-performing model.

Final Performance

91.50% Test Accuracy

0.91 Macro F1

The model demonstrated particularly strong recognition of Glioma and Meningioma, while Pituitary tumor classification remained the more challenging category because of its lower recall.

---

12. Medical Physics and Future Development

This project was developed at the intersection of medical imaging, deep learning, and quantitative analysis , with relevance to the growing role of computational methods in modern medical physics.

The current model should be considered a research/educational classification system rather than a clinical diagnostic tool . High classification accuracy alone is not sufficient for clinical deployment.

The next stage of the project focuses on:

• Confusion-matrix analysis
• Sensitivity and specificity
• ROC-AUC analysis
• Precision-recall analysis
• False-positive and false-negative investigation
• Grad-CAM explainability
• Visualization of image regions influencing model predictions
• Model limitations and potential clinical considerations
• Development of a small Streamlit demonstration interface

Grad-CAM and error analysis are particularly important because they can help investigate whether the CNN is responding to medically meaningful regions of the MRI rather than irrelevant image characteristics.

---

Conclusion

This project demonstrated the application of transfer learning and deep convolutional neural networks to brain MRI tumor classification .

Four architectures were systematically evaluated under a consistent experimental framework. While DenseNet121 and ResNet50 provided useful comparative baselines, fine-tuned VGG16 achieved the best performance with 91.50% test accuracy and a 0.91 macro F1-score .

The project also demonstrated that successful medical-image deep learning requires more than selecting a powerful CNN. Preprocessing, architecture selection, classification-head design, layer freezing, fine-tuning depth, learning rate, optimization and regularization all influence model performance.

The final stage will focus on model interpretability and clinically relevant evaluation , moving the project beyond simple accuracy-based classification toward a more complete medical-imaging research workflow.