Contextual Importance and Utility (CIU) for explaining image classification. This py.ciu.image package implements CIU for explanation of image classification results.
CIU was developed by Kary Främling in his PhD thesis Learning and Explaining Preferences with Neural Networks for Multiple Criteria Decision Making, (written in French, title Modélisation et apprentissage des préférences par réseaux de neurones pour l'aide à la décision multicritère), available online for instance here: https://tel.archives-ouvertes.fr/tel-00825854/document. It was originally implemented in Matlab and has later been re-implemented in R and Python (packages ciu and py-ciu) for tabular data.
This package was fundamentally modified in October 2023 to make it general-purpose (rather than being completely specific for the study of bleeding in gastro-enterological images).
From PyPI by doing pip install py_ciu_image, or e.g. by cloning this repository for the latest (or other) version.
A Jupyter notebook is provided that shows how to produce CIU explanations for Imagenet networks (VGG16 and Resnet-152 are used). The notebook is CIU_Imagenet.ipynb.
Another Jupyter notebook goes through how gastroenterological images can be "explained" with CIU for bleeding and non-bleeding images, as reported in the paper Explainable artificial intelligence for human decision support system in the medical domain. The notebook is CIU_gastro_images.ipynb. The original 3295 images in the Red Lesion Endoscopy data set are publicly available and can be retrieved from Coelho https://rdm.inesctec.pt/dataset/nis-2018-003, last accessed on 26 October 2023.
The package has been documented using Sphinx and is available at https://py-ciu-image.readthedocs.io/. Sphinx HTML documentation can also be generated by doing cd docs and then make html. Other formats such as PDF can also be produced, as for the PDF version at: docs/py_ciu_image.pdf.
The initial R implementation is accessible at (https://github.com/KaryFramling/ciu.image) and was also used in the gastroenterological paper from 2021. However, it is unlikely that the R package would be further developed due to the lack if deep neural network implementations in R and that the transfer of images between R and python is slow.
- Kary Främling: CIU "inventor", programmer of the initial ciu.image code in "R" and of the total re-factoring and fixing of the package in Oct/Nov 2023.
- Vlad Apopei: Finalisation, programming and packaging of py.ciu.image.
- Rohit Saluja: Programming of the initial implementation of py.ciu.image.
- Manik Madhikermi: Support and testing.
- Avleen Malhi: Initiator and co-ordinator of medical image explanation work.