Abstract
This is the era of Artificial Intelligence, where AI models are growing fast for making critical decisions for several predictive models. Here AI is giving solution to the medical practices for diagnosis as well as prognosis with its increasing intelligence to simplify the ambiguity and complexity in data to carry out clinical decisions. Several research studies have truly demanded the need of AI-based systems and how to enhancing their capabilities to help medical practitioners. However, instead of giving highest effort for making most accurate AI based models, still now assessing the magnitude and impact of human trust on AI technology demands substantial attention. In the last decade many AI based CAD models were developed which hardly could persuade the experienced medical practitioners to accept the machine-specified decisions. In this research work, it was attempted to interpret and explain a supervised AI model built using XGBoost on Lung cancer detection by using XAI (Explainable AI) tools - two post-hoc methods (LIME and SHAP) and one ante-hoc method, to provide satisfactory explanations to medical practitioners, thereby minimize the AI risk factor in implementation of the model and reinforce the trust to the medical experts and patients in accepting such model. In this paper, the results of all three XAI tools were illustrated using heatmaps to select important input bi- omarkers that contributed more in detection of the benign or malignancy state of the pulmonary nodules. Finally the supervised AI model was rebuilt using only those important input features and it was found out that the metrics like specificity, precision, the AUC of the newer model under the ROC curve were giving better result in prediction of lung cancer nodule state. It is a study to explore how XAI tools highlight the contributions of input features in an AI model and how that AI model’s performance can be fine-tuned based on the outputs of XAI mechanism.
Keywords: AI; LIME; Pulmonary nodules; SHAP; XAI; XGBoost
References
- AMA passes first policy recommendations on augmented intelligence.
- An introduction to explainable AI with Shapley values.
- The Global Burden of Disease 2004 Update, WHO.
- Hancock, M., Pylidc , MIT.
- Rathe A. Random Forest vs XGBoost vs Deep Neural Network, Kaggle.
- Zhu P and Ogino M. “Guideline-based additive explanation for computer-aided diagnosis of lung nodules”. In Interpretability of Machine Intelligence in Medical Image Computing and Multimodal Learning for Clinical Decision Support (2019): 39-47.
- Singh A, Sengupta S and Lakshminarayanan V. “Explainable deep learning models in medical image analysis”. Journal of Imaging 6.6 (2020): 52.
- Lucieri A., et al. “Achievements and Challenges in Explaining Deep Learning based Computer-Aided Diagnosis Systems”. arXiv preprint (2020).
- Ahmed ZU., et al. “Explainable artificial intelligence (XAI) for exploring spatial variability of lung and bronchus cancer (LBC) mortality rates in the contiguous USA”. Scientific reports 11.1 (2021): 1-15.
- Siddhartha M, Maity P and Nath R. “Explanatory artificial intelligence (XAI) in the prediction of post-operative life expectancy in lung cancer patients”. Int J Sci Res 8 (2020).
- Bartczak M and Partyka M. Chapter 8 Story Lungs: eXplainable predictions for post operational risks.
- Venugopal VK., et al. “Unboxing AI-radiological insights into a deep neural network for lung nodule characterization”. Academic radiology 27.1 (2020): 88-95.
- Zhou B., et al. “Learning deep features for discriminative localization”. In Proceedings of the IEEE conference on computer vision and pattern recognition (2016): 2921-2929.
- Selvaraju RR., et al. “Grad-cam: Visual explanations from deep networks via gradient-based localization”. In Proceedings of the IEEE international conference on computer vision (2017): 618-626.
- Lundberg SM and Lee SI. “A unified approach to interpreting model predictions”. In Proceedings of the 31st international conference on neural information processing systems (2017): 4768-4777.
- Shrikumar A, Greenside P and Kundaje A. “Learning important features through propagating activation differences”. In International Conference on Machine Learning (2017): 3145-3153.
- The Cancer Imaging Archive (TCIA) Public Access, LIDC-IDRI.
- Tulio Ribeiro M, Singh S and Guestrin C. "Why Should I Trust You?": Explain- ing the Predictions of Any Classifier. arXiv e-prints (2016): arXiv-1602.
- He H., et al. “ADASYN: Adaptive synthetic sampling approach for imbalanced learning”. In 2008 IEEE international joint conference on neural networks (IEEE world congress on computational intelligence) (2008): 1322-1328.