변수 중요도를 이용한 설명 가능한 인공지능 기법의 시각화에 대한 고찰과 보건정보 자료에의 응용
Visualization of Explainable Artificial Intelligence Techniques Using Variable Importance with Its Applications to Health Information Data
1
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
Objectives: Deep learning techniques have been actively used in the medical field where precise diagnosis and results are very important. Deep neural network-based models utilizing big data from medical records are supporting medical opinions and are revolutionizing the medical industry. In addition, the convolutional neural network model shows excellent performance in analyzing image data and are used for image classification and X-ray/CT image reconstruction. Methods: In this paper, we conducted a visualization study using structured and unstructured data in the medical field. Results: In order to determine input variables affecting mortality and to evaluate their importance, a total of five techniques, namely, the augmented neural network model with multi-task learning, random forest, extra tree, gradient boosting and xgboost are applied to the intensive care unit data. Variable importance is calculated for each technique, and these indicators are all converted to ratios in consideration of the differences considering the patient group as a stratification variable. The converted values are shown in three graphs, a lollipop graph, a bubble chart graph, and a heat map graph. Through the visualization, it was easy to see which variables were relatively important for each technique and to what extent. InceptionResnetV2 was used as a classification model for skin cancer image data, and LIME and Grad-CAM were applied to the model to easily identify the characteristics of each cancer. Conclusions: Through this study, we apply several explainable artificial intelligence techniques to medical data to enhance understanding of the results of analysis and to help identify and visualize important input variables and features.
Abstract
Objectives: Deep learning techniques have been actively used in the medical field where precise diagnosis and results are very important. Deep neural network-based models utilizing big data from medical records are supporting medical opinions and are revolutionizing the medical industry. In addition, the convolutional neural network model shows excellent performance in analyzing image data and are used for image classification and X-ray/CT image reconstruction. Methods: In this paper, we conducted a visualization study using structured and unstructured data in the medical field. Results: In order to determine input variables affecting mortality and to evaluate their importance, a total of five techniques, namely, the augmented neural network model with multi-task learning, random forest, extra tree, gradient boosting and xgboost are applied to the intensive care unit data. Variable importance is calculated for each technique, and these indicators are all converted to ratios in consideration of the differences considering the patient group as a stratification variable. The converted values are shown in three graphs, a lollipop graph, a bubble chart graph, and a heat map graph. Through the visualization, it was easy to see which variables were relatively important for each technique and to what extent. InceptionResnetV2 was used as a classification model for skin cancer image data, and LIME and Grad-CAM were applied to the model to easily identify the characteristics of each cancer. Conclusions: Through this study, we apply several explainable artificial intelligence techniques to medical data to enhance understanding of the results of analysis and to help identify and visualize important input variables and features.
참고문헌 (17)
- Olden JD, "An accurate comparison of methods for quantifying variable importance in artificial neural networks using simulated data", Ecol Model, 178(3-4), pp.389~397 (2004)[DOI]
- Alain G, "Understanding intermediate layers using linear classifier probes", arXiv preprint 2018;arXiv:1610.01644, ~
- Samek W, "Explainable artificial intelligence:understanding, visualizing and interpreting deep learning models", arXiv preprint 2017;arXiv:1708.08296, ~
- Explainable artificial intelligence (XAI), "Gunning D.", Defense Advanced Research Projects Agency, ~ (2017)
- Ribeiro MT, "“Why should I trust you?” Explaining the predictions of any classifier", Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp.1135~1144 (2016)
- Selvaraju RR, "Grad-cam: visual explanations from deep networks via gradient-based localization", Proceedings of the IEEE International Conference on Computer Vision, pp.618~626 (2017)
- Feng J, "Nonparametric variable importance using an augmented neural network with multi-task learning", International Conference on Machine Learning, pp.1496~1505 (2018)
- Lamy JB, "Explainable ar-tificial intelligence for breast cancer : a visual case-based reasoning approach", Artif Intell Med, 94, pp.42~53 (2019)[DOI]
- Sabol P, "Explainable classifier supporting decision-making for breast cancer diagnosis from histopathological Images", 2019 International Joint Conference on Neural Networks, pp.1~8 (2019)
- Zhou B, "Learning deep features for discriminative localization", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp.2921~2929 (2016)
- Selvaraju R, "GradCAM: visual explanations from deep networks via gradient-based localization", proceedings of the IEEE International Conference on Computer Vision, pp.618~626 (2017)
- Silva I, "Predicting in-hospital mortality of ICU patients: the physionet", computing in cardiology challenge 2012, 2012 (2012), 245~248
- Pedregosa F, "Scikitlearn : machine learning in Python", JMLR, 12(85), pp.2825~2830 (2011)
- Soo Y, "Affects of white blood cell count for cardiovascular disease", pp.2006
- Wickham H, "ggplot2: elegant graphics for data analysis" (2016), Springer-Verlag
- Tschandl P, "The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions", Sci Data, 5, pp.180161~ (2018)[DOI]
- Gregorutti B, "Grouped variable importance with random forests and application to multiple functional data analysis", Comput Stat Data Anal, 90, pp.15~35 (2015)