Skip to content

Latest commit

 

History

History
70 lines (37 loc) · 2.07 KB

README.md

File metadata and controls

70 lines (37 loc) · 2.07 KB

Interpretation of 3D CNNs for Brain MRI Data Classification

We analyze data provided by The Human Connectome Project (HCP). Using T1-weighted MRI data we solve the problem of binary classification of task of gender patterns recognition between men and women with 3D - CNN. Further we intepret obtained model to undestand of gender-related brain differencies.

Please refer to the sourse paper.

Setup and Dependencies

Install all dependencies with

pip install -r ./requirements.txt

DATABASE

The data we use is an open-access database taken from Human Connection Project (HCP). We worked with morphometry description of T1 MPI images as wel as the full-sized images preprocessed in Freesurfer according to the HCP pipeline.

Data contain 1113 subjects, including 507 men and 606 women. Each object is represented by a 1 GB ZIP archive with a name corresponding to a unique object ID. Each archive contains a lot of information. For automatic access to the target MRI file, the power shell script was written that can be found in

./data/DATE_ACCESS.md

The script allows you to extract the necessary file from the internal ZIP archive(inside the main archive), without unzipping the main one. Also, a unique ID corresponding to each object is assigned as a name for each file.

Masks

To obtain all masks use

./masks/obtain_masks.ipynb

CNN Model

To train the 3D CNN models use

./model3d/training_model.ipynb

Architecture reproduced from the paper Brain Differences Between Men and Women: Evidence From Deep Learning

Meaningful perturbation

3D visualization of mask

Interpretation with meaningful perturbation:

GradCAM

Interpretation with Grad CAM:

Guided backpropagation

3D CNN Interpretation with Guided backpropogation: