BrainScan AI
Brain Tumor Detection AI
A personal deep learning project to explore TensorFlow and Keras capabilities in medical image analysis
Project Overview
BrainScan AI is a personal project aimed at learning and mastering TensorFlow and Keras through practical application. The project focuses on developing a Convolutional Neural Network (CNN) for automated brain tumor classification from MRI scans. This self-directed learning experience helped me gain hands-on experience with deep learning frameworks while tackling a meaningful medical imaging challenge.
Learning Objectives
- Master TensorFlow framework and Keras API
- Understand CNN architecture design and optimization
- Apply deep learning to medical image analysis
- Learn model evaluation and performance metrics
Technical Details
Key Features
- 94% overall classification accuracy
- Multi-class tumor classification
- Custom CNN architecture
- Optimized for 256x256 MRI scans
Model Performance
Confusion Matrix
Detailed classification results across tumor types
Classification Accuracy by Type
- Glioma: 93.6% accuracy
- Meningioma: 87.7% accuracy
- No Tumor: 98.0% accuracy
- Pituitary: 96.7% accuracy
Technical Implementation
Model Architecture
Model: Sequential
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
rescaling (Rescaling) (None, 256, 256, 3) 0
conv2d (Conv2D) (None, 256, 256, 16) 448
max_pooling2d (MaxPooling) (None, 128, 128, 16) 0
...
dense_1 (Dense) (None, 4) 516
=================================================================
Training Progress
Model accuracy and loss evolution during training
Example Predictions
Sample predictions across different tumor types
Confidence Analysis
Distribution of model confidence for correct predictions
Complete Implementation Details
Below is the complete implementation of the brain tumor detection model, including data processing, model architecture, training, and evaluation.