MIDM: Feature structured interpretable XGBoost network for breast cancer
Dehua Chen, Keting Zhong, Jianrong He · 2021 International Conference on Computer Information Science and Artificial Intelligence (CISAI) · 2021
The incidence rate of breast tumors is the first place in women’s malignant tumors, which seriously threatens human health and life. The accuracy of traditional AI diagnosis model based on machine learning is not high; the accuracy of modern AI diagnosis model of breast cancer based on deep learning is improved, but the interpretability of diagnosis process is insufficient, and doctors are not easy to trust. In this paper, we propose a Mammography Interpretable Diagnosis Model (MIDM), which is based on the construction of medical text semantic tree to achieve structured, and combined with XGBoost model to achieve breast cancer text classification. At the same time, the Local Interpretable Model Agnostic Interpretation (LIME) method is used to explain the prediction results of the example. Our model was evaluated based on the real mammography dataset, which was significantly better than other comparison models, and the interpretable characteristics were roughly the same as the clinicians’ consideration characteristics in distinguishing benign from malignant.