VoxCNN for ADNI 3D Brain MRI classification (ISPR final project in UNIPI)

Model architecture :

      VoxCNN (Voxels convolutional neural network) model architecture

How to run the code ?

1. Prerequisites :

  nvidia-docker run -v "Your local data set path":/root -it gcr.io/tensorflow/tensorflow:latest-gpu bash
  pip install --upgrade nibabel
  pip install --upgrade keras==2.1
  pip install --upgrade setGPU

2. Set data set and log path :

  ds_path = Your data set path
  plot_history_path = Your log path

3. Start training :

  python ADNI.py

4. Calculate results and plotting :

  python Plot.py

Result 1:

1. ROC AUC plot for AD vs Normal classification of VoxCNN

2. ROC_AUC : [mean] ± [std]

AD vs NC : [0.87538462] ± [0.06169398]
AD vs EMCI : [0.65433333] ± [0.05580634]
AD vs LMCI : [0.76464286] ± [0.06621509]
LMCI vs NC : [0.66934524] ± [0.05654292]
LMCI vs EMCI : [0.51897321] ± [0.15433192]
EMCI vs NC : [0.49821581] ± [0.0859987]

3. evaluation acc : [mean] ± [std]

AD vs NC : [0.74743083] ± [0.1340696]
AD vs EMCI : [0.62246155] ± [0.11438376]
AD vs LMCI : [0.5398693] ± [0.08587749]
LMCI vs NC : [0.53092732] ± [0.10495938]
LMCI vs EMCI : [0.66389989] ± [0.01105777]
EMCI vs NC : [0.5365353] ± [0.04308016]

Result 2:

1. ROC AUC plot for AD vs Normal classification of VoxCNN

2. ROC_AUC : [mean] ± [std]

AD vs NC : [0.91935897] ± [0.0777566]
AD vs EMCI : [0.633] ± [0.03178312]
AD vs LMCI : [0.71464286] ± [0.07319599]
LMCI vs NC : [0.71561355] ± [0.07265784]
LMCI vs EMCI : [0.57989583] ± [0.10804529]
EMCI vs NC : [0.51877137] ± [0.09879062]

3. evaluation acc : [mean] ± [std]

AD vs NC : [0.71185771] ± [0.07849011]
AD vs EMCI : [0.66184616] ± [0.05658861]
AD vs LMCI : [0.61895427] ± [0.05988868]
LMCI vs NC : [0.60912281] ± [0.04352362]
LMCI vs EMCI : [0.60981556] ± [0.10904755]
EMCI vs NC : [0.53706441] ± [0.07779821]