Brain Scan Symmetry Anomaly Detection

This project uses an autoencoder to detect anomalies in brain scans by measuring the symmetry between the left and right hemispheres. The model is trained in an unsupervised manner to identify brain anomalies, such as tumors or lesions, based on asymmetries in the reconstructed scans.

Methodology

• The autoencoder is trained on healthy brain scans, learning to reconstruct symmetrical images.
• Asymmetry between the reconstructed hemispheres signals an anomaly.
• A contrastive loss function is used to compare the left and right brain embeddings, detecting abnormal scans.

Requirements

Install the required libraries:

pip install tensorflow keras opencv-python numpy matplotlib scikit-image seaborn scipy sklearn

Usage

1. Prepare Dataset:

   • Place 200 brain CT scans (100 normal, 100 abnormal) in the head_ct/ folder.
   • The images should be grayscale and named 001.png, 002.png, etc.

2. Run Training and Testing:

   • Train the autoencoder and test it by running:

     python train_autoencoder.py

   • This script trains the model on normal brain scans, then automatically tests the model on both normal and abnormal scans, producing results like ROC curves and AUC values.