- This repository contains all necessary code to replicate the experiments in Deep Neural Networks or Dermatologists?: http://arxiv.org/abs/1908.06612.
- We suggest to start with train_and_test_set_creation.ipynb.
- Then train a suite of models using the DataSplit_HpSearch.py file.
- Then produce Kernel SHAP and GradCAM explanations: "Shap_GradCAM_Notebooks".
- Sanity check code is contained in "Model_Sensitivity_Experiments" and "Randomised_layer_experiments"
Arxiv: http://arxiv.org/abs/1908.06612
Sample Bibtex file:
@InProceedings{10.1007/978-3-030-33850-3_6,
author="Young, Kyle
and Booth, Gareth
and Simpson, Becks
and Dutton, Reuben
and Shrapnel, Sally",
editor="Suzuki, Kenji
and Reyes, Mauricio
and Syeda-Mahmood, Tanveer
and Glocker, Ben
and Wiest, Roland and Gur, Yaniv
and Greenspan, Hayit
and Madabhushi, Anant",
title="Deep Neural Network or Dermatologist?",
booktitle="Interpretability of Machine Intelligence in Medical Image Computing and Multimodal Learning for Clinical Decision Support",
year="2019",
publisher="Springer International Publishing",
address="Cham",
pages="48--55",
isbn="978-3-030-33850-3" }
- Rethinking the Inception Architecture for Computer Vision (CVPR 2016)
- Tschandl, P., Rosendahl, C., Kittler, H.: The ham10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Scientific data 5, 180161 (2018)
- Lundberg, S.M., Lee, S.I.: A unified approach to interpreting model predictions. In: Advances in Neural Information Processing Systems. pp. 4765–4774 (2017), https://github.com/slundberg/shap
- Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., Batra, D.: Gradcam: Visual explanations from deep networks via gradient-based localization. In: Proceedings of the IEEE International Conference on Computer Vision. pp. 618–626 (2017)
- Kotikalapudi, Raghavendra: keras-vis, https://github.com/raghakot/keras-vis