Introduction

Brain Titer prediction is usually intrusive and only possible after medical inspection. This is an attempt to perform prediction of few of those titers using Deep Learning algorithms.

Brain Abeta Prediction

This attempts to improve performance of Abeta prediction by using Inception network and a VGG-inspired lighter convolution based network as backbone of GlobalLocal Transformer.

Train the models

cd Brain\ Abeta\ Estimation/Abeta\ train
python GLT_train_test.py

Validate the models

cd Brain\ Abeta\ Estimation/Abeta\ train
python GLT_Trial_incep.py
python GLT_Trial_novel.py

Additional Regression Data Analysis

Brain\ Abeta\ Estimation/Regression\ Try\ 1.ipynb

Brain Age Prediction

This presents detailed code and results for lobe-wise fusion models and Global Local Transformer performance for Brain-Age prediction.

Fusion Network

Preprocessing brain MRI

cd Brain\ Age\ Estimation/Fusion\ Network/Preprocessing
python script.py
python flirt.py
python ravens.py

Preprocessing lobewise segments Training Fusion Network

cd Brain\ Age\ Estimation/Fusion\ Network/Preprocessing
python mask.py

Training Fusion Networks

cd Brain\ Age\ Estimation/Fusion\ Network/train
python trainreg.py

Testing Fusion Networks:

cd Brain\ Age\ Estimation/Fusion\ Network/train
python test.py

The above code is to train for the entire brain. Use the lobe-wise training codes for part based training.

GLT Network: GLT was also explored for intensity and ravens map of brain MRI scans.

Training Scripts:

cd Brain\ Age\ Estimation/Global\ Local\ Transformer/train
python train_test_val.py
python train_test_val_ravens.py

Test Scripts:

cd Brain\ Age\ Estimation/Global\ Local\ Transformer/train
python test_script.py
python test_script_ravens.py

Acknowledgements

This project is forked from FiANet and uses Global Local Transformer.