Computer aided diagnosis semantic model for the report of medical image via LDA and LSA

In this paper, we build a probabilistic model used to represent medical text data which contents latent semanteme. We manage “hidden topic” with Latent Dirichlet Allocation (LDA) respectively, and use Latent Semantic Analysis (LSA) to construct the semantic structure. The medical text data that we use is reports of medical image which include two parts: image description and diagnoses. Traditional task demand expert to diagnose illness according to the image description and his/her own subjective assumptions whose accuracy is based on expert´s experience, and bring more workload. Here we present models that require no manual diagnosis and also automatically give the diagnoses with high readability. The learning models can be used for both classification and analysis of medical text data.