Medical Case-Based Retrieval of Patient Records Using the RadLex Hierarchical Lexicon

We present a new method for the retrieval radiological cases from a database of clinical cases described by terms from the RadLex lexicon. The input is an database of cases and a query consisting of the patient volumetric scan, a user-defined region of interest in it, and a list of RadLex from the radiological report. The output is list of the most relevant cases from the database in decreasing order. Our method uses the RadLex terms and their hierarchical representation to define a similarity metric between terms based on their relative location in the hierarchy. For this purpose, we develop the Augmented RadLex Graph, a data structure that augments the RadLex hierarchy with links derived from the terms in the case reports, and a search algorithm that ranks case similarity based on the link distance between the terms in the graph. Our method was evaluated in the VISCERAL Retrieval Benchmark Challenge on 8 queries and a database of 1,813 cases. It ranked first in 6 out of the 8 cases tested.