https://github.com/borgwardtlab/tda-cnn-ad
Model combining topological descriptors with patch based MR imaging features
https://github.com/borgwardtlab/tda-cnn-ad
Last synced: about 1 year ago
JSON representation
Model combining topological descriptors with patch based MR imaging features
- Host: GitHub
- URL: https://github.com/borgwardtlab/tda-cnn-ad
- Owner: BorgwardtLab
- Created: 2020-11-13T10:25:34.000Z (over 5 years ago)
- Default Branch: main
- Last Pushed: 2020-11-13T15:05:35.000Z (over 5 years ago)
- Last Synced: 2025-01-22T04:14:01.281Z (over 1 year ago)
- Language: Python
- Size: 23.4 KB
- Stars: 0
- Watchers: 3
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# TDA-CNN-AD
Model combining topological descriptors with patch based MR imaging features.
This is a work in progress repositroy by F. Hensel and S. Brueningk according to the initial description in https://arxiv.org/abs/2011.06531. This code will be further imporved and hence may sightly deviate from the original description.
In out analysis we used T1-weighted MR images for AD and CN subjects from the Alzheimer's Disease Neuroimaging Initiative ([ADNI](http://adni.loni.usc.edu)). Data was preprocessed as described in the archive article using the [fmriprep](https://github.com/nipreps/fmriprep) pipeline. For the creation of persistence images, we first calculated the persistence diagrams of the full MRIs using [dipha](https://github.com/DIPHA/dipha) and then subsequently computed the persistence images using [persim](https://github.com/scikit-tda/persim).
The function run.py contains the code to run the image-patch-based 3D-CNN, the TDA 2D-CNN, and a combined model using both topoligical descriptors and a 3D image patch. The information for all 216 patched can be combined in a logistic regression model (ensemble model 1), whereas the preclassification layer encodings of the TDA 2D-CNN and single patch 3D-CNN can used as features for a single dense layer (ensemble model 2).