| Summary: | 碩士 === 國防醫學院 === 公共衛生學研究所 === 106 === The Classification of Disease are used in a wide range of applications. This classification task is typically routinely performed by professional clinical coders in hospital. However, shortage of qualified/trained coders decreased accuracy of payments.
Automated disease classification system is studied in many researches. Classify of disease are usually represented following the bag-of-words (BoW) method. Despite being one of the simple and effect way used representations in disease classification, the BoW model has its own limit.
In our work, using word2vec to carry more semantic meanings. For the classification disease task, we propose a method combining word embedding with a convolutional neural network (CNN).
Multichannel with two sets of word vectors and fine-tuning during training, had best performance in our study. Weighting the loss function make us successfully to identify third-level ICD-10-CM higher prevalence codes. This shows that future studies have to overcome the class imbalance problem.
|
|---|
