Diagnosis labeling with disease-specific characteristics mining

Data analysis and management of huge volumes of medical data have attracted enormous attention, since discovering knowledge from the data can benefit both caregivers and patients. In this paper, we focus on learning disease labels from medical data of patients in Intensive Care Units (ICU). Specifically, we extract features from two main sources, medical charts and notes. We apply the Bag-of-Words (BoW) model to encode the features. Different from most of the existing multi-label learning algorithms that take correlations among diseases into consideration, our model learns disease specific features to benefit the discrimination of different diseases. To achieve this, we first construct features specific to each disease by conducting clustering analysis on its positive and negative instances, and then perform training and testing by querying the clustering results. Extensive experiments have been conducted on a real-world Intensive Care Units (ICU) database. Evaluation results have shown that our proposed method has better performance against all other compared multi-label learning methods.