Raghava-Ram/brain-tumor-efficientnet

library_name: tensorflow tags:

• —medical-imaging
• —brain-tumor
• —mri
• —image-classification
• —deep-learning
• —efficientnet ---

Brain Tumor MRI Classification (EfficientNetB4)

Model Description

This model performs automatic classification of brain MRI scans into four categories using deep transfer learning. It is built using the EfficientNetB4 architecture pretrained on ImageNet and fine-tuned for brain tumor classification.

The model predicts one of the following classes:

• —Glioma Tumor
• —Meningioma Tumor
• —Pituitary Tumor
• —No Tumor

The objective of this project is to assist in automated analysis of brain MRI images and demonstrate how deep learning can be applied to medical imaging tasks.

Model Details

Architecture: EfficientNetB4 + custom classification head Framework: TensorFlow / Keras Input size: 380 × 380 RGB images Output: Softmax probabilities for 4 classes

Custom classification head:

EfficientNetB4 (ImageNet pretrained)
BatchNormalization
Dense(256) + ReLU
BatchNormalization
Dropout(0.4)
Dense(128) + ReLU
BatchNormalization
Dropout(0.3)
Dense(4) + Softmax

Dataset

The model was trained on a brain MRI dataset containing images categorized into four classes.

Training Set

• —Total images: 5,600
• —Classes: 4
• —Images per class: 1,400

Testing Set

• —Total images: 1,600
• —Images per class: 400

Classes:

• —Glioma
• —Meningioma
• —Pituitary
• —No Tumor

All images were resized to:

380 × 380 pixels

Training Procedure

Training was performed in two phases.

Phase 1 – Transfer Learning

• —EfficientNetB4 backbone frozen
• —Only classification layers trained
• —Optimizer: Adam
• —Learning rate: 1e-3

Phase 2 – Fine Tuning

• —EfficientNetB4 backbone unfrozen
• —Entire network trained
• —Learning rate reduced to 1e-4

Callbacks used:

• —EarlyStopping
• —ReduceLROnPlateau

Maximum training epochs:

40 + 40 epochs

Evaluation

The model was evaluated on a separate test dataset.

Test Accuracy: ~95%

Evaluation metrics included:

• —Accuracy
• —Confusion Matrix
• —Classification Report

Usage

Download Model

Model repository:

https://huggingface.co/Raghava-Ram/brain-tumor-efficientnet

Direct model file:

https://huggingface.co/Raghava-Ram/brain-tumor-efficientnet/resolve/main/pretrained_model.keras

Load the Model

from huggingface_hub import hf_hub_download
import tensorflow as tf

model_path = hf_hub_download(
    repo_id="Raghava-Ram/brain-tumor-efficientnet",
    filename="pretrained_model.keras"
)

model = tf.keras.models.load_model(model_path)

Intended Use

This model is intended for:

• —Educational purposes
• —Research demonstrations
• —Medical imaging deep learning experiments
• —AI model deployment projects

Limitations

• —The dataset size is limited.
• —Performance may vary across different MRI machines or imaging conditions.
• —The model has not been clinically validated.

Ethical Considerations

⚠️ This model is for educational and research purposes only.

It should not be used for real medical diagnosis or treatment decisions. Always consult qualified medical professionals for clinical interpretation.

Author

Raghava Ram

Project: Brain Tumor Detection using Deep Transfer Learning