Research And Implementation Of Multi-label Learning Based On The Cases Of Chronic Hepatitis B

The rapid development of hospital information system has produced a large amount of medical information, which makes historical electronic medical records and patient’s transaction data become complicated, so it becomes a hot spot to get useful information from massive clinical data in medical information research. As a branch of data mining, classification has been widely used in many areas. Traditional classification usually assumes that instances are associated with only one label. However, instances are usually associated with multiple labels such as video, text, images, disease diagnosis and treatment recommendation. Apparently, traditional classification algorithms are not applicable in these areas. Therefore, the study of multi-label data becomes an important research topic. This paper introduced the theory and framework of multi-label learning, and extracted the dataset of Chronic Hepatitis B from the databases of the hospital information system. It also obtained the experiment results for proposed algorithms in the dataset of Chronic Hepatitis B, and the performance between the new method and the old ones was compared and analyzed. At last, we implemented the combination treatment recommendation system.Firstly, the framework and strategy of multi-label classification were elaborated in this paper. Then related algorithms about multi-label learning were summarized and the evaluation indexes of multi-label learning were introduced. Secondly, Chronic Hepatitis B patient’s basic information, initial lab information and clinical medical information were extracted from the EMR database, HIS database and LIS database of a hospital, which provided the dataset for experiment. Then, we reduced the dimensionality of the feature information of the dataset and made mapping for the dataset with some typical multi-label algorithm. After each medicine has been marked with a tag, multi-label dataset could be preprocessed, including data cleaning, data integration and data reduction, etc. On this basis, the paper proposed the LIFT-ML-KNN algorithm. Using this algorithm and the old ones, we trained the dataset of Chronic Hepatitis B successfully and compared and analyzed several key evaluation results. Finally, we implement the prediction of the combination treatment with java language in MyEclipse, which provided a strong support for clinical path of Chronic Hepatitis B.