Gifsplanation via Latent Shift: A Simple Autoencoder Approach to Counterfactual Generation for Chest X-rays

Joseph Paul Cohen, Rupert Brooks, Sovann En, Evan Zucker, Anuj Pareek, Matthew P. Lungren, Akshay Chaudhari
Proceedings of the Fourth Conference on Medical Imaging with Deep Learning, PMLR 143:74-104, 2021.

Abstract

Motivation: Traditional image attribution methods struggle to satisfactorily explain predictions of neural networks. Prediction explanation is important, especially in medical imaging, for avoiding the unintended consequences of deploying AI systems when false positive predictions can impact patient care. Thus, there is a pressing need to develop improved models for model explainability and introspection. Specific problem: A new approach is to transform input images to increase or decrease features which cause the prediction. However, current approaches are difficult to implement as they are monolithic or rely on GANs. These hurdles prevent wide adoption. Our approach: Given an arbitrary classifier, we propose a simple autoencoder and gradient update (Latent Shift) that can transform the latent representation of a specific input image to exaggerate or curtail the features used for prediction. We use this method to study chest X-ray classifiers and evaluate their performance. We conduct a reader study with two radiologists assessing 240 chest X-ray predictions to identify which ones are false positives (half are) using traditional attribution maps or our proposed method. Results: We found low overlap with ground truth pathology masks for models with reasonably high accuracy. However, the results from our reader study indicate that these models are generally looking at the correct features. We also found that the Latent Shift explanation allows a user to have more confidence in true positive predictions compared to traditional approaches (0.15$\pm$0.95 in a 5 point scale with p=0.01) with only a small increase in false positive predictions (0.04$\pm$1.06 with p=0.57). Accompanying webpage: https://mlmed.org/gifsplanation/ Source code: https://github.com/mlmed/gifsplanation

Cite this Paper

BibTeX
@InProceedings{pmlr-v143-cohen21a, title = {Gifsplanation via Latent Shift: A Simple Autoencoder Approach to Counterfactual Generation for Chest X-rays}, author = {Cohen, Joseph Paul and Brooks, Rupert and En, Sovann and Zucker, Evan and Pareek, Anuj and Lungren, Matthew P. and Chaudhari, Akshay}, booktitle = {Proceedings of the Fourth Conference on Medical Imaging with Deep Learning}, pages = {74--104}, year = {2021}, editor = {Heinrich, Mattias and Dou, Qi and de Bruijne, Marleen and Lellmann, Jan and Schläfer, Alexander and Ernst, Floris}, volume = {143}, series = {Proceedings of Machine Learning Research}, month = {07--09 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v143/cohen21a/cohen21a.pdf}, url = {https://proceedings.mlr.press/v143/cohen21a.html}, abstract = {Motivation: Traditional image attribution methods struggle to satisfactorily explain predictions of neural networks. Prediction explanation is important, especially in medical imaging, for avoiding the unintended consequences of deploying AI systems when false positive predictions can impact patient care. Thus, there is a pressing need to develop improved models for model explainability and introspection. Specific problem: A new approach is to transform input images to increase or decrease features which cause the prediction. However, current approaches are difficult to implement as they are monolithic or rely on GANs. These hurdles prevent wide adoption. Our approach: Given an arbitrary classifier, we propose a simple autoencoder and gradient update (Latent Shift) that can transform the latent representation of a specific input image to exaggerate or curtail the features used for prediction. We use this method to study chest X-ray classifiers and evaluate their performance. We conduct a reader study with two radiologists assessing 240 chest X-ray predictions to identify which ones are false positives (half are) using traditional attribution maps or our proposed method. Results: We found low overlap with ground truth pathology masks for models with reasonably high accuracy. However, the results from our reader study indicate that these models are generally looking at the correct features. We also found that the Latent Shift explanation allows a user to have more confidence in true positive predictions compared to traditional approaches (0.15$\pm$0.95 in a 5 point scale with p=0.01) with only a small increase in false positive predictions (0.04$\pm$1.06 with p=0.57). Accompanying webpage: https://mlmed.org/gifsplanation/ Source code: https://github.com/mlmed/gifsplanation} }
Endnote
%0 Conference Paper %T Gifsplanation via Latent Shift: A Simple Autoencoder Approach to Counterfactual Generation for Chest X-rays %A Joseph Paul Cohen %A Rupert Brooks %A Sovann En %A Evan Zucker %A Anuj Pareek %A Matthew P. Lungren %A Akshay Chaudhari %B Proceedings of the Fourth Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2021 %E Mattias Heinrich %E Qi Dou %E Marleen de Bruijne %E Jan Lellmann %E Alexander Schläfer %E Floris Ernst %F pmlr-v143-cohen21a %I PMLR %P 74--104 %U https://proceedings.mlr.press/v143/cohen21a.html %V 143 %X Motivation: Traditional image attribution methods struggle to satisfactorily explain predictions of neural networks. Prediction explanation is important, especially in medical imaging, for avoiding the unintended consequences of deploying AI systems when false positive predictions can impact patient care. Thus, there is a pressing need to develop improved models for model explainability and introspection. Specific problem: A new approach is to transform input images to increase or decrease features which cause the prediction. However, current approaches are difficult to implement as they are monolithic or rely on GANs. These hurdles prevent wide adoption. Our approach: Given an arbitrary classifier, we propose a simple autoencoder and gradient update (Latent Shift) that can transform the latent representation of a specific input image to exaggerate or curtail the features used for prediction. We use this method to study chest X-ray classifiers and evaluate their performance. We conduct a reader study with two radiologists assessing 240 chest X-ray predictions to identify which ones are false positives (half are) using traditional attribution maps or our proposed method. Results: We found low overlap with ground truth pathology masks for models with reasonably high accuracy. However, the results from our reader study indicate that these models are generally looking at the correct features. We also found that the Latent Shift explanation allows a user to have more confidence in true positive predictions compared to traditional approaches (0.15$\pm$0.95 in a 5 point scale with p=0.01) with only a small increase in false positive predictions (0.04$\pm$1.06 with p=0.57). Accompanying webpage: https://mlmed.org/gifsplanation/ Source code: https://github.com/mlmed/gifsplanation
APA
Cohen, J.P., Brooks, R., En, S., Zucker, E., Pareek, A., Lungren, M.P. & Chaudhari, A.. (2021). Gifsplanation via Latent Shift: A Simple Autoencoder Approach to Counterfactual Generation for Chest X-rays. Proceedings of the Fourth Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 143:74-104 Available from https://proceedings.mlr.press/v143/cohen21a.html.

Related Material