Cassandre Project

Made by : Matthieu Bachelot

What is Project Cassandre ?

• Master thesis project
• Use deep learning methods on MRI volumes
• Detect tumors on MRI slice using semantic segmentation

Methodology

• Test several architectures
• Test several data modeling / data generation methods
• Select the best

Architectures

• Auto Encoder / PSPnet / UNet / SegNet
• Kepp their architectures intact
• Test metric / loss / optimizers

Data generation techniques

Ressources constraints

Data modeling methods

Ground Truth representation

Final results

Focus on two networks -> reduce test time

• Statistical study on test data
• PSPNet was the best overall
• Training using Mini batch generation / One hot Vector representation / Data Augmentation

Limits

• Lot of result variation between patients
• Heterogeneous dataset
• time limits