Semantic rule-based determination of cancer stages from free-text radiology reports

Abstract

Cancer staging provides a basis not only for determining proper treatments for a patient but also for making future national-wide health plans. Despite its benefits, it is very difficult to obtain staging data because it is not commonly performed for all cancer patients or simply not collected. Moreover, it requires medical experts to do the analysis, incurring expensive cost. In this paper, we propose a method for semantic rule based determination of a cancer stage, which is considering a semantic type (e.g. body part, organ, or organ component). Compared to previous work, our work is unique in that we utilize radiology reports instead of pathological report since they are more available. Moreover, we argue that a rule-based approach is more suitable for cancer staging than machine learning because the international staging protocols specify certain conditions for determining stages. Since semantic type of words should be considered to determine the cancer stage and construct rules, we construct rules using MetaMap, which provides a meta-thesaurus of UMLS (Unified Medical Language System) for medical text. Based on our semantic rules, TNM (Tumor Nodes Metastasis) stages are determined for 275 reports of liver cancer. From our experiments, whole performance are highly incremented than machine learning approach.